Some time in October last year, a company landed in our pipeline that my partner Ricardo flagged as “fairly outside of AI :)”. As we found out, there’s actually quite a bit of AI in it — but it’s certainly not a typical software company. What Aardaia wants to do is something humans have barely done in the last few thousand years: create an entirely new agricultural crop.

My first reaction was a combination of “very interesting!” and “but food is tough…”, but after my first call with Pádraic Flood and Mike Henske, the founders of Aardaia, I posted “LAFS!!!” (short for “Love at first sight”) to our Zendesk. Fast forward a few weeks, and we led the company’s seed round.

In this post I’ll try to explain what made us so bullish about the company. But first, some background.

When I started digging in, I knew next to nothing about the seed business (despite having been a seed investor for almost two decades). My biology knowledge is limited to faint, 30-year-old memories of high-school biology and a long-standing amateur interest in evolution, shaped largely by Richard Dawkins’ fantastic books. But I greatly enjoyed learning a little about the underlying biology, and wanted to share some of what I learned.

Here’s a question … what are some new crops that have been introduced in modern history? As I learned, the list is surprisingly short. Soy is probably the best example: an ancient Asian crop that became globally relevant only in the 20th century. Everything else — wheat, rice, corn, potatoes, cassava, beans — was domesticated over centuries or millennia of patient selection.

The reason is that domesticating a new crop is incredibly slow: you typically get one generation per growing season, and each generation nudges a trait only a little. Turning a wild plant into something farmable means stacking up dozens of traits at once — yield, uniformity, taste, easy harvesting, disease resistance — and historically that took hundreds or thousands of generations.

So what Pádraic and Mike set out to do is compress something that used to take millennia into less than a decade. And they’re doing it without gene modifications.

That is the biological problem they’re trying to solve. The commercial opportunity they’re going to address is that the world, and Europe in particular, needs more, better plant protein. Demand for clean, functional plant proteins keeps climbing — the market is on track to roughly double, from around $19B today to $40B over the next decade. But every option on the market has a catch. Soy is cheap, but it’s imported and comes with flavor and sustainability baggage. Pea protein tends to taste off and doesn’t gel well. Potato protein works beautifully — white, neutral, gels great — but there isn’t much of it and it’s expensive.

It’s a global issue, but Europe feels it especially. The continent produces plenty of calories, but it still imports a large share of the high-protein plant feed used for livestock, especially soy from the Americas. That is inefficient, bad from a sustainability point of view, and increasingly a political problem, with growing pressure to produce more protein at home.

What Aardaia is building

Aardaia is domesticating the aardaker (Lathyrus tuberosus), a long-forgotten European legume that grows edible tubers. If you’ve never heard of it, same here. The closest thing in my head was “aardvark”.

The aardaker turns out to be a pretty unusual plant:

1. Aardaker can work with nitrogen-fixing bacteria to convert nitrogen from the air into a form the plant can use, reducing the need for synthetic fertilizer. What’s more, unlike soy, which usually has to be “inoculated” with compatible bacteria because its bacterial partners are not reliably present in European soils, aardaker’s compatible rhizobia are much more likely to already be present in local soils.

2. Unlike potatoes, whose tubers are mostly starch and relatively low in protein, aardaker stores significant protein directly in its tubers (around 25% of dry matter), white, bland, and strongly gelling.

3. It was actually grown in the Netherlands before the Industrial Revolution, then lost out to the potato.

So nature already spent hundreds of millions of years evolving a plant with more or less exactly the properties the protein market is asking for. It just never got turned into a real crop, because until recently that was impossibly slow. The thesis is that you can combine potato-like yields, soy- or pea-like protein, and potato-like starch into one new European crop.

Aardaia doesn’t introduce any foreign genes and doesn’t modify the genome using modern gene-editing techniques. It’s good old-fashioned breeding — pick the best plants, cross them, repeat — just enormously sped up. Instead of inventing new traits or trying to transfer traits from other species, they’re searching through a huge, barely touched pool of natural genetic variation for the good combinations that already exist, and putting them together.

Why is that pool so rich? Crops like maize and wheat have been bred for ~10,000 years, and along the way they’ve lost most of their genetic variation. They’re close to their ceiling — squeezing another 1% of yield out of them is a big deal. The aardaker has barely been touched, so almost all of its natural variation is still sitting there, unused. Pádraic likes to point out that their greenhouse — full of wild plants collected from about 20 countries — probably holds more genetic diversity than the entire human species. (We’re all unusually similar to each other genetically, because humanity squeezed through a population bottleneck around 200,000 years ago. The aardaker never did.) More variation to play with means more room to keep improving the crop, year after year.

How it works

This is roughly how they turn all that raw variation into an actual crop, fast:

1. They built a high-quality, chromosome-level reference genome of one single plant (or one specific genotype, to be more precise). Every other plant they sequence gets lined up against that reference, which lets them see exactly where it differs. (Later you can blend several plants into a “pan-genome” so you’re not leaning on just one individual, but you start with a single good reference.)

2. Genomic prediction. They grow tens of thousands of different plants, connect DNA differences to traits (yield, protein, flavor, disease resistance), and build models that predict which crosses will produce offspring better than anything they’ve grown so far. At the seedling stage, a cheap DNA test from a leaf punch tells them which plants are worth keeping — so a lot of the selection happens on a computer before the field does the slow, expensive part. Cattle breeders, for example, have used the same statistical approach for years. What’s new is pointing the full modern toolkit at domesticating a brand-new crop — which nobody has done, and which wasn’t even affordable until 5–10 years ago, when sequencing costs fell by roughly a thousandfold.

3. Speed breeding. In climate-controlled chambers, they push plants to flower early. Their fastest have gone from seed to flower in about five weeks. The goal is a ~10-week seed-to-seed cycle, which would mean up to five generations a year instead of the one you’d get in a field — a 5–10x speedup of evolution itself.

4. Shuttle breeding. They grow in multiple hemispheres and climates at the same time (with seed orchards in the Netherlands and in Sardinia), which speeds up adaptation and shows early on how the crop behaves in different soils and weather.

Most of the world is trying the opposite approach: take an existing crop and engineer a new trait into it. The Gates-funded effort to engineer nitrogen fixation into cassava, for example, is a roughly 30-year project. Aardaia’s bet is that it’s faster and cheaper to find a plant that already has the trait you want and breed it into a crop. The aardaker already fixes nitrogen and already makes protein-rich tubers. As one plant scientist put it to me: a better crop with no fertilizer, versus the same old crop with a bit less.

Plenty of companies use genomics and modern breeding — but to improve crops we already have (a better potato, a better wheat). Almost nobody else is trying to genomically domesticate a genuinely new one. That gives Aardaia a head start — and the lead compounds, because they keep piling up proprietary plant material, a hybrid-seed system, and a genotype-to-trait dataset nobody else has.

The key questions

There are obviously still risks and unknowns.

1. Biology and yield. Can they get field-average yields up fast enough? Today they’re at around 4 t/ha. ~10 t/ha gets them level with potato protein, ~40 t/ha level with soy, and 50+ t/ha cheaper than soy. Their best individual plants already extrapolate to well over 30 t/ha — but a single star plant isn’t the same as a whole field’s average, and closing that gap across seasons, soils, and diseases is the hard, slow part. What makes me optimistic is that the crop’s huge untapped variation, plus the speed-breeding loop, give Aardaia far more shots on goal per year than a normal breeder gets. And the early seasons are already trending up.

2. Agronomy. Since the aardaker hasn’t been farmed at scale in about 150 years, the basic farming manual — spacing, ridging, fertilizer, irrigation, harvesting — has to be written almost from scratch. The good news is that these are knowable, adjustable variables, not deep scientific mysteries.

3. Market and adoption. Farmers won’t plant a crop with no buyers, and processors won’t commit without supply. Here the demand is already real and pulling: Aardaia sells every tuber it can grow to top restaurants today, and it’s in active talks with large European food-ingredient players. The restaurants build the brand and the story while the ingredient business gets built underneath.

This needs amazing talent, knowledge, and engineering ingenuity. But it does not need a miracle. It needs roughly an 8x improvement in yield — hard, but not the kind of 100x leap some deep-tech bets require — and it already makes money today. Aardaia sells its tubers to a celebrated two-Michelin-star plant-based restaurant in the Netherlands, where the limit is supply, not demand. Every field they plant pulls double duty: it brings in revenue, and it produces the ~100,000-plant dataset that feeds the next round of breeding.

While the near-term business is restaurants and ingredients, the long-term business is built on hybrid-seed genetics, which makes it look more like biotech, pharma — or software. When two carefully selected parent lines are crossed, the first-generation crop can have exceptional traits, but those traits do not reliably reproduce in the next generation. So the IP protection is built into the biology itself — farmers buy fresh seed every season because that is the only way to reliably get the high-performing crop.

When we talked to people who know Pádraic, the same things kept coming up: that he is one of the most impressive plant scientists of his generation; that this crop is his life’s mission; and — my favorite, that he is “pure”, someone who genuinely wants to help and means it. He did a PhD in plant genetics, spent five years at the Max Planck Institute, and was Director of Genetics at Infarm before this. Mike is the commercial and operational counterpart, bringing deep agribusiness experience. Together, they grew the aardaker program about fivefold with basically just the two of them, before raising a cent.

If you’ve ever wanted to work where genomics, AI, and the physical world of farming meet, Aardaia is hiring across plant breeding, computational biology and genomics, AI for breeding, agronomy, and food science … go talk to Pádraic and Mike!

Last week I wrote that we, a firm known mostly for its focus on B2B SaaS, have invested in startups working on autonomous navigation for spacecraft, personalized cancer vaccines, and micro-drones that hunt mosquitoes (stealth), among other companies that significantly expand the scope of what software investing meant traditionally.

Today, I’m excited to share that we participated in Anthropic’s Series G.

A VC that has “early stage” encoded in its freaking *name* and that prides itself on the craft of seed investing, participating in a financing at a $380 billion valuation?! You might wonder if we’ve lost our minds.

How it happened

The opportunity came about when our portfolio company Vercept was acquired by Anthropic in January/February, and we got the chance to not only roll over our position but double down. The timing made the decision easier than it might sound. Anthropic had just disclosed that its revenue run-rate had grown from $1B to $9B over the course of 2025, an unprecedented pace at this scale. We’d also observed that since the release of Opus 4.5 and 4.6, developers had been switching to Claude in droves, and it helped that we experienced the capability jumps firsthand. On top of that, there were rumors about a much more capable model in the pipeline (since announced as Mythos).

What does this mean for our strategy?

We’re not turning Point Nine into a late stage fund. And we’re not changing our name to two point zero. ;-) We continue to focus on early stage investing. This is an exception, because of a special opportunity to invest in a one-of-a-kind company.

In early April, Anthropic announced it had crossed $30 billion in annualized revenue. A few weeks later, it reportedly crossed $40B. As crazy as it sounds, Anthropic’s revenue would be even higher if the company had more compute, since the market for the highest-intelligence tokens is supply-constrained. To say that these numbers are “exceptional” would be an understatement. It’s unheard of in the history of capitalism.

What better company to make an exception for?

Over the last few years, something happened though. A sizable portion of the companies we invest in these days deal not only with bits but with atoms too, so to speak (and sometimes photons, molecules, and proteins).

In the last two years, we’ve partnered with startups working on:

• 200-ton autonomous dump trucks for open-pit mines (Sensmore)

• Personalized cancer vaccines (Serova)

• Autonomous navigation for spacecraft (EraDrive)

• Micro-drones that hunt mosquitoes (stealth)

• A nitrogen-fixing, protein-rich crop that doesn’t exist in agriculture today (stealth)

Not your father’s SaaS portfolio. ;-)

We still invest in software, but in many cases it’s software and AI for the world outside of offices:

• Hula Earth’s on-device AI identifies nearly 10,000 animal species, giving landowners a real-time picture of the biodiversity on their land.

• Draxon uses VR to train airport ground handling crews.

• Sereact is teaching warehouse robots to pick, place, and sort objects they’ve never seen before.

• Rerun is building the data infrastructure for robotics and computer vision.

• Upciti uses sensors and cameras to provide cities with real-time data to optimize operations.

• (Stealth) is building the operating system for weather modification.

Where we still invest in pure software, it’s mostly foundation models or agentic systems that require extraordinarily deep domain knowledge to build, such as:

• Vercept (foundation model for computer use, recently acquired by Anthropic)

• Poolside (foundation model for software agents)

• Forithmus (foundation models for medical imaging)

• Findable (AI buiding intelligence for real estate)

• Zauber (AI agents for sea and air freight forwarders)

• (Stealth) (training methods for deeply superhuman coding LLMs)

A lot of these companies are working on things that could have an enormous impact on the world, using technologies that would have been impossible to build just a few years ago, which makes our jobs as early-stage investors more interesting than ever.

If we try to put them on a map (because that’s what VCs do :) ), we can see that the majority of our companies live at the intersection of “Engineered World and AI” or “Nature and AI”:

Of course there’s no perfect way to do this. You can argue some companies should be moved around a little or you can pick different dimensions … but we like this way of looking at it because it shows the three major themes we’ve been gravitating toward over the last few years.

Many of the above products are delivered as-a-service, so often it’s still a SaaS business model, albeit with different pricing, so not everything we’ve learned about SaaS has become irrelevant. But as you can see, what we’re looking for in a software company today vs. some years ago has (not surprisingly) changed quite dramatically.

2010-style SaaS investing is dead. Software investing has never been more exciting! (*)

(*) I realize this ending sounds like ChatGPT slop. I swear I wrote it myself!

The strongest published benchmarks include Zapier’s hiring rubric, Fin’s 2x productivity program, Ramp’s Glass platform, and Jobber’s eight-rung engineering ladder, which was inspired by this blog post. They agree that AI fluency is observable, repeatable, and measured against output, not adoption alone.

This index condenses that consensus into 6 levels (L0–L5) based on 13 questions across 5 dimensions (Culture, Talent & Workflows, Throughput, Tooling & Context, Product, and Accountability & Governance). It shouldn’t take more than 10–15 minutes to complete and gives you an overall score as well as detailed report card with more granular scores and recommendations.

Head over to www.ai-pilled.com and let me know how you’ve scored! :)

How AI-pilled are you?

A bunch of people have asked me about the status of my Knowledge Hub, which I wrote about a few weeks ago. How well does it work? Did I get PMF inside Point Nine? Things are changing so fast that I’ve been hesitant to write this post … by the time I hit the “publish” button, things might look different again. ;-) But I owe y’all an update, so here goes.

TL;DR

I’ve moved to a much simpler and probably better solution. Instead of worrying about a vector database, embeddings, RAG, hybrid search, a custom MCP server, and a process supervisor, I’ve just connected our data sources (Attio, Slack, Zendesk, etc.) directly with Claude, and it seems to work just fine.

A Quick Recap

About a month ago, I vibe-coded a Knowledge Hub: ~43,800 lines of Python, six data source connectors, a vector database, MCP servers, and more. I wrote about it here and also shared this (AI-generated) tech doc.

It was promising. It sort of worked. But it kept breaking. Connections got lost, MCP servers went down, sync jobs failed in non-obvious ways. Some sources kept failing silently. The app would happily report that it had synced, but nothing had actually made it into the database. Other times, documents were in the database but for some reason weren’t being retrieved. Plus all kinds of little things that didn’t work.

All of this is fixable, and better AI coding models and tooling come out almost on a daily basis, making it easier. But while working on it, I’ve tried a much simpler approach in parallel … and realized that the previous approach was heavily over-engineered (for our use case). It also helped that Anthropic released improvements at breakneck speed, including more plugins and better connections to GMail/G-Drive/G-Cal.

Doing this, I realized that by simply plugging everything into C̶l̶a̶u̶d̶e̶ Claude Code directly, I could probably get 90–95% of the value for about 10% of the effort. Especially 10% of the maintenance effort … and that matters a lot. A simpler AI Command Center system (Claude suggested this title when I discussed the specs with it) means I can iterate faster, because there’s so much less complexity to manage.

Side-by-Side: Knowledge Hub vs. AI Command Center

Here’s a quick comparison of the two systems (mostly written by Claude):

The main advantage of the Knowledge Hub is speed, because it provides the AI with a pre-indexed search. With the AI Command Center, question answering is pretty slow because everything happens live. But as mentioned before, the latency advantage of the Knowledge Hub comes with quite a lot of operational complexity. If it turns out that the AI Command Center isn’t good enough from a speed and answer quality perspective, I might go back to the Knowledge Hub — or take a closer look at systems like Glean or Dust.

New Features!

Because the system was suddenly so much lighter, I could experiment more freely. So I started adding things:

Task management. I made it my task manager to capture ideas and to-dos, replacing what I’d previously used ChatGPT for (which I wrote about before). The main advantage vs. the ChatGPT solution is that ideas and tasks are now saved in a Notion database, which is much more robust than relying on the context window of ChatGPT.

Smart screenshot workflows. When I paste a screenshot of a LinkedIn profile, it triggers research on that founder and their company. When I paste a screenshot of a WhatsApp conversation about scheduling, it triggers a Blockit scheduling workflow.

Morning digest. Every morning, it compiles a summary — emails, Slack messages, calendar updates — so I can start the day with a quick overview. I don’t know yet how useful it will actually be, but hey, why not? It took almost no effort to add.

But the most exciting feature I’ve added is a super simple but pretty interesting “scout”, which works surprisingly well so far, where Claude is instructed to find interesting founders/companies that I should take a look at. It looks for interesting themes in various places, does a first round of research, saves them as leads and presents them to me.

With this feature, I ran into a pretty serious issue: researching companies involves a large number of tool calls, and this kept clogging up the context window to the point where Claude would just stop working. I tried a bunch of approaches to slim it down. Nothing worked well enough … until I switched from Claude Chat to Claude Code for this workflow. The key difference isn’t that Claude Code has a larger context window (it doesn’t). The big advantage (no news to you if you’re an engineer) is that Claude Code can spawn sub-agents — farming out parts of a complex task to separate agents, each with their own fresh context window. This way, the parent agent’s context doesn’t balloon.

Since it works much better for the scouting/research use case and since I frequently run into context size issues with various types of questions, I’ve just migrated the entire thing to Claude Code (I told you there’s a chance that it might change before I’m done with this post).

And what about the Crustacean Revolution?!

The obvious question some of you might have: Why am I not using OpenClaw for this? Good question. For now, I’m moving pretty fast with what I have, but I’m experimenting with OpenClaw in parallel, and it’s well possible that I’ll soon switch to a more agentic, OpenClaw-powered setup … I’ll keep you posted!

Imagine you’re an early-stage software company and you’ve just negotiated one of your first contracts with a large European enterprise customer. The customer had a lot of extra requests and wanted a large discount. You hesitated, weighing the risks (esp. the engineering capacity tied up by one customer and the danger that what you build won’t fully align with your roadmap) against the upside of signing a reference customer with a strong logo that can unlock future sales. To tip the balance, you insisted on a three-year term, which the customer accepted.

You might think: Great — all fair and square, both sides got what they wanted.

Well, no — at least not according to the EU.

Under the new EU Data Act, your customer will have the right to terminate the contract on two months’ notice, irrespective of the three-year term you agreed(1). If you think that sounds crazy, it’s because it is. Protecting consumers from being locked into unfair contracts is one thing. Overriding carefully negotiated, fixed-term agreements between sophisticated commercial parties is another.

Where this all comes from

According to the EU, the termination right is a necessary consequence of another principle codified in the new law: the right to switch to another data processing service(2). This switching right seems fair. If, after signing a long-term contract, a customer finds another supplier with much better service or pricing, it shouldn’t be tied-in simply because the existing supplier refuses to cooperate.

What’s surprising is why the right to switch should necessitate another right altogether — the termination of the existing contract. The EU argues that termination is necessary to make the switching right effective. This may well be the case for a small retail customer subscribing to a mass-market cloud service with standard, non-negotiable terms. Yet most consumer-oriented software already offers short-term plans anyway, often due to national fairness rules for standardized terms.

For the many long-term contracts individually negotiated between professional parties, this rationale does not apply. Yes, it’s fair and common sense that a customer should be able to request transfer of its data at any time, but that has nothing to do with honouring the underlying contract when it was negotiated and agreed at eye level.

The “proportionate” enigma

The EU seems to understand this, having added the following sentence to the recitals of the law (recital 89):

“Nothing in this Regulation prevents a customer from compensating third-party entities for support in the migration process or parties from agreeing on contracts for data processing services of a fixed duration, including proportionate early termination penalties to cover the early termination of such contracts, in accordance with Union or national law.” (highlighting by the author)

Setting aside the contradiction between a fixed-term contract and a general termination right, this sentence raises more questions than it answers. What does “proportionate” mean? Proportionate to what? What does “in accordance with Union or national law” mean, given that national law generally respects fixed-term contracts as a natural part of contractual freedom and therefore requires full payment of the agreed fees even if a customer no longer wants the service? How could anything be considered “disproportionate” if it follows the basic principle of pacta sunt servanda?

The only sensible answer is that “proportionate” does not limit the contractual freedom of consenting professionals and that contractually agreed payment obligations must be honoured — even if a party decides to switch to another provider and no longer wants the service. But if that is the case, why stipulate a termination right at all instead of simply regulating the switching right? Why introduce an ominous qualifier like “proportionate” in relation to remaining fees, creating massive legal uncertainty, rather than focusing on the supplier’s cooperation obligation?

There’s no obvious explanation, at least none that we can see. The best hope at this point is for the EU to quickly clarify its intention, and for practitioners, first and foremost legal advisers, not to speculate about its meaning but to stick to established legal principles (pacta sunt servanda). If you decide to take this route, you’re in good company. There are plenty of cloud-based service providers that demand full payment, often pointing to upfront costs and significant discounts granted in return for long-term revenue certainty (see e.g. Asana, Cisco, Cloudflare, IONOS, Salesforce, teradata).

All this notwithstanding, it may be worth considering a somewhat more conciliatory approach, especially if you’re a smaller provider without a massive legal budget. After all, the “termination right cum proportionate penalties” now exists. Even if it is pointless and contrary to basic legal principles, accommodating it in a workable way might appease counterparties and — if push comes to shove — make it easier for a judge to decide in your favour.

If you want to follow this route, we’d suggest allowing for a small reduction of fees for the time after termination. A reasonable benchmark could be your approximate gross margin or the costs that fall away if the service no longer needs to be provided (also referred to as “saved expenses,” e.g. third-party cloud fees). To keep things manageable, setting termination fees somewhere between 80% and 95% of the original contract fees should be fully defensible.

Other “goodies”

The termination right is by far the most consequential provision of the Data Act for data processors, but it’s not the only one to keep in mind. Here are some of the other “goodies” the Act has in store:

• If you thought you might be out of scope because you’re small or don’t materially change customer data, forget it. The Data Act definition is extremely wide. The moment you provide “centralized or distributed computing resources,” you’re in, regardless of company size. The only way around the Data Act might be if you install your software on servers exclusively owned or run by the customer (“on-premises”) without any third-party access.

• You’re not just subject to the switching right; you must also include it in every customer contract — including contracts concluded long before the Data Act. How to do this without customer consent remains an open question.

• You need to add extensive additional content to contracts, including an “exhaustive” list of all data and digital assets that can be ported (Art. 25).

• You must inform customers about switching and porting methods, formats, restrictions, and technical limitations, and publish an online register detailing available data structures and formats (Art. 26).

• Your website must inform customers about the jurisdiction governing your data-processing infrastructure and the measures taken to protect non-personal data against “international governmental access” (Art. 28).

As of today, only a few member states have enacted enforcement legislation for the Data Act (e.g. the Netherlands, Finland, Denmark, and Malta). Yet many more are in the making, making it advisable to implement the necessary changes sooner rather than later, even if you’re not located in an early-adopter country. To illustrate how contractual matters could be handled in practice, we found the addenda published by AWS and Cisco particularly helpful.

Fingers crossed

In a best-case scenario, the required changes to contract terms and websites will be a one-time effort and won’t materially impact the economics of long-term software contracts. In that case, the Data Act would mean more red tape, but not a major burden for the European startup scene. If, however, a more far-reaching interpretation of the “proportionate” limitation prevails or causes lasting legal uncertainty, European data processing providers may face yet another competitive disadvantage compared to rivals outside the EU, who don’t have to deal with the Data Act as long as they don’t serve EU customers. Fingers crossed.

Notes: (1) The EU also requires you to inform your customer about the termination right and make it part of the contract with you, even if the contract dates back to a time before the Data Act came into force (12 September 2025).

(2) Somewhat strangely the Data Act does not expressly stipulate the termination right, but rather seems to assume it’s pre-existent: Art. 23 obliges data processing providers to implement the switching right (which triggers the termination of the contract) in customer contracts and make certain information available (Art. 25, 26) but it remains unclear where the (assumed) switching/termination right comes from.

13 days and ~43,800 lines of Python later, here’s an update.

What it does

Knowledge Hub (the not very creative working title) pulls data from the primary systems that we run on at Point Nine: Gmail, Google Drive, Slack, Zendesk, Attio, and Granola.

The system can also ingest data from ChatGPT. Given how many conversations we’ve all had with ChatGPT over the last few years, I thought this might make sense, but it’s still TBD how useful this will be in practice. So this is an experimental feature in an already experimental application. The Dropbox integration has been built but not yet enabled in production. Since we use Dropbox to store sensitive documents, I want to invest more time into security testing before enabling it.

All data gets processed into a vector database (Qdrant) with semantic search. It’s then connected to Claude via MCP servers (local for Claude Desktop, remote for Claude.ai in the browser). Using Claude, you can ask natural-language questions like:

• “How is Company X doing?”

• “Have we seen Company Y?”

• “What did we discuss with Company Z?”

…and get AI-powered answers with sources.

In addition, I’ve also created a Slack bot that allows you to get information from the vector database in a faster way. It uses regex-based question classification and only hits Claude for the final answer (with tighter constraints — shorter responses, no extended thinking). This makes it snappier for quick lookups (what you expect from a Slack bot) while Claude with MCP gives you Claude’s full reasoning power for a larger variety of questions.

Architecture

At a high level, the system is split into three main layers.

• At the top sits a simple web app that handles authentication, user management, and the dashboard. This is the part users interact with directly.

• Below that is a set of background services responsible for syncing data from external systems, running scheduled jobs, and exposing the knowledge base via MCP servers and a Slack bot.

• At the bottom are the data sources themselves and the vector database, where all content is ingested, embedded, and queried.

If you’re interested in more technical details, here is an (AI-generated) technical documentation.

Key Features

• Hybrid search — combines semantic (meaning-based) and keyword search

• Multi-user support — Google OAuth login; each user connects their own accounts

• Role-based access — admin vs. user permissions

• Claude integration — MCP servers let Claude Desktop and Claude.ai search the knowledge base directly

• Slack bot — team members can query the knowledge base from Slack

• Automated sync — scheduled background jobs keep data fresh

275 commits. 95 pull requests. Zero lines written by a human.

As you probably guessed, I didn’t write a single line of code myself. Every line was written by either Replit’s agents or Claude Code. I described what I wanted, tested the result, and told the AI what wasn’t working (many, many times).

Besides that, I occasionally asked the AI for a full code review and for suggestions for improvements, with a lot of rinse and repeat.

I did have to deal with pull requests, merges, version conflicts, etc., mostly because I messed up things when I tried to have several AIs working on it in parallel, and because it quickly went way out of my technical comfort zone. Because I don’t know how a system like this has to be built, the iteration went much slower than it would have if an experienced software developer had done it.

Things I’ve struggled with

• Process supervision. The MCP servers kept crashing. It took a lot of back and forth to implement health checks, auto-restart logic, and process monitoring. Neither I nor the AI anticipated this upfront.

• Parallelization. Initial sync was painfully slow. We were processing everything sequentially. (Am I starting to refer to the AI and me as “we,” a team?) Once we added parallel processing and batching, we saw a big speedup.

• Edge cases. All kinds of little things that created bugs that had to be fixed, in most cases without me even looking into what was going on.

I’m still battling with several issues: the MCP servers occasionally go down, the dashboard sometimes shows inconsistent numbers, sync jobs fail in non-obvious ways, and the Granola import is messy. I’m pretty confident that we (yes, we!) will be able to fix all of this though.

What’s less clear is how good the system will ultimately be at answering non-trivial real-life questions. Based on my tests so far, it provides pretty impressive answers on tasks like “Write a memo about company X” or ”How is company Y doing?”. But once we (I mean my human colleagues and I) start using it seriously — asking more sophisticated questions across more diverse topics — it’s well possible that it won’t be good enough to be genuinely useful.

My expectation is that my AI and I will be able to implement a few more relatively easy wins, but that pushing quality meaningfully beyond that will be hard. So my guess is that we’ll eventually end up using a product built by someone who knows what they’re doing. :-)

Build vs. Buy

This is not an argument against vibe coding or against customization. Keep in mind that (i) I’m not an engineer and I did this part-time over two weeks, (ii) what I’ve tried to build here is a somewhat complex system, and iii) RAG across large, messy, heterogeneous datasets is not a solved problem. What I’ve built only scratches the surface of what’s needed to do this well. Finding needles in haystacks — reliably and verifiably — requires much more: better chunking strategies, reranking, evaluation frameworks, feedback loops, domain-specific tuning…

So in this particular case (small company, complex system, difficult problem), SaaS is still the better choice. In many other cases, the answer will be different, which leads us to the trillion-dollar question:

What kind of software will be replaced by custom-built solutions?

Given the speed at which AI coding models are improving, I’m careful with predictions, but my current thinking is:

1) For a small company like ours: If the solution exists, we should buy SaaS. It doesn’t make sense to reinvent the wheel (unless you do it for the fun and the learning, like me).

2) For a large enterprise: For a company paying millions of dollars for SaaS and with internal technical resources, the build-vs-buy calculus changes. I have strong conviction that if not now, then in the near future, a small team of engineers with AI assistance will in many cases be able to build and maintain a custom solution for a fraction of the cost of traditional SaaS. If I look at what I as an amateur have managed to build in less than two weeks, I can only imagine what a good developer, directing several AI coding agents in parallel, can get done in a few months.

I don’t think the replacement of SaaS by custom-built applications will happen fast. For most companies, reducing IT/software spend is not a top priority. If a piece of software works well, replacing it just to save money will usually not be a priority. It’s different when the existing software doesn’t do a great job and I think that’s where the “let’s just build it” conversations will start. But at some point, inevitably, the question will be asked more frequently: “Why are we spending $2 million on this and $500k on that?”

Now, software companies obviously have access to the same (or even better) tools as those enterprises. And arguably they should be able to leverage AI even more. What I mean by that is, if SaaS is under threat because software development becomes 10–100x cheaper, maybe the solution for SaaS companies is to produce 10–100x more software to keep the pendulum where it is.

Some more thoughts on vibe coding

When we talk about vibe coding, we’re actually talking about two developments:

1. Non-developers building simple software. People using Lovable, Replit Agent, or Bolt to build simple applications that they wouldn’t have been able to build before. Landing pages, internal tools, basic automations.

2. Engineers becoming hyper-productive. For complex systems, you’ll still need engineers. I built a 43,000+ LOC system without writing code, but I also spent 13 days debugging issues that an experienced engineer would have anticipated. I believe engineers paired with AI are becoming unbelievably productive. One person doing what used to take a team. Small teams doing what used to take hundreds of people.

If you’re a non-technical person thinking about building something with AI assistance and you’re overwhelmed or intimidated: just start. Start on Replit or Lovable. Just ask the AI what to build and iterate. If you get error messages, paste them into the AI chat. If you get stuck, ask Claude Opus 4.5 for help. If you don’t know how to give Claude access to the code, ask ChatGPT or Claude how to do it. You’ll get a good response, and in many cases, the AI will not only give you the answer but can do it for you straight away. You’ll be amazed by how far you can get. LFG!

In these earlier explorations, we argued that their operational playbook(s) and value proposition(s) often diverge meaningfully from traditional software businesses and that their addressable markets could often be multiple times larger.

What wasn’t clear to us was how these businesses could balance automation and service quality, how they could scale operations without linear cost growth, how they could build strong (data) moats, or how they could articulate value beyond cost savings.

After dozens of founder conversations, deep dives into emerging sub-verticals, and some early commercial signals from the market, we now have more evidence to revisit those questions. As part of this, we’re also sharing an updated version of the market map to reflect where we’re seeing the most activity and momentum.

What Mapping Verticals Revealed About Workflow Structure

We began by mapping verticals to understand where AI-first service businesses are most likely to succeed. The initial pattern suggested that the most promising opportunities lie in under-digitized industries with entrenched, low-NPS incumbents like property management. But what’s proven even more important is the shared structure of the workflows these businesses aim to replace.

Whether it’s insurance claims, tax filings, property management, or immigration law, the underlying work is often highly structured and repeatable, driven by documents, rules, and checklists rather than creative problem-solving. Most of it still runs on PDFs, spreadsheets, and legacy systems.

That shared anatomy makes these categories particularly well-suited to being rebuilt from the ground up:

• Structured intake and triage: In insurance or property management, each case starts with a flood of semi-structured inputs, e.g., KYC packs, maintenance tickets, scanned forms. AI-native firms use LLM + OCR pipelines to classify and validate this data automatically, routing only edge cases to humans.

• Document-heavy review and cross-referencing: Tax firms or claims processing specialists often rely on junior/lower-qualified staff to search through dense statutes, policy docs, or precedent letters. RAG and vector search now surface the relevant clause in seconds.

• Repeatable, rules-based decisioning: Whether it’s approving a claim, closing an alert, or filing a compliance opinion, outcomes often follow fixed logic. AI-first teams train policy agents to apply those rules, explain the outcome, and log it immutably.

• Low-value manual tasks at scale: Across verticals, teams still spend hours copying data across systems, reconciling ledgers, or filling out forms. End-to-end automation reduces marginal cost per case to near zero and scales throughput without linear hiring.

• Massive untapped historical data: Legacy firms sit on decades of returns, claims, filings, and alerts, mostly untouched. AI-native companies structure this data to train vertical-specific models that improve accuracy and create defensibility.

What emerges is a new class of service business: data-in, judgment-out factories. They run on documents, structured data, and rules-based logic. The output is a decision, classification, or filing, increasingly handled by agents instead of human analysts.

Since our original market map, we’ve seen a noticeable uptick in activity across insurance-related services, e.g., Inca or Elysian in claims processing and Flow or Meshed in brokerage. Financial services remain a strong category, with steady expansion across tax, accounting, and compliance. And we’re beginning to see more unique and complex use cases emerge, like Convexia in drug discovery or Operand in management consulting.

Revisiting the Open Questions

Some things weren’t clear early on, i.e., how far automation could go without hurting quality, what scalable delivery would look like, or how these businesses would differentiate beyond price. We’re starting to see more answers take shape.

1/ Getting to the Right Automation Level Without Compromising Service Quality

Across most verticals we’ve looked at, companies are not aiming for full automation from day one. Instead, they’re building workflows where AI handles the bulk of routine tasks and humans remain involved at key points. Specifically in:

• Regulatory compliance: In many verticals, there are legal ceilings on automation. German customs brokers are legally required to manually review filings. Property managers must conduct in-person meetings annually. Tax advisors often need certified oversight. In these contexts, human accountability isn’t optional, and will not become so in the near future.

• Accuracy assurance: In high-stakes workflows (claims, filings, tax, security), automation errors are costly. Some of the companies we’ve talked to invest heavily in custom verification layers, reviewer training, and tightly controlled workflows. Control sheets, QA loops, and task-specific overrides ensure that speed doesn’t come at the expense of accuracy.

• Trust and relationship management: Some industries are still fundamentally human, e.g., brokers, real estate agents, wealth advisors. These customers often care more about trust and service than technical elegance. Integral, an AI-native tax advisory for German SMBs, doesn’t mention AI once on its homepage. Their customers aren’t looking for sophistication; they’re looking for confidence.

We’re particularly excited about companies that manage to productize parts of their service early, without rushing into full automation too soon, or defaulting to stitching together off-the-shelf tools without real leverage. Getting automation right is less about maximizing coverage and more about sequencing it properly, starting with the aspects of the business that provide the biggest operational leverage.

2/ Balancing scalability vs. growth

In AI-first services, growth without automation just means more people. And more people likely lead to margin compression, coordination risk, and brittle operations. We’ve been excited to see some companies starting out by building the systems that allow margins to expand with volume by encoding expertise into infrastructure.

Offdeal calls this “the platform”, which effectively is a unified foundation of structured data, live context, and agent interfaces. Others, like one AI-native TPA in insurance we’ve spoken with, have embedded years of adjuster know-how into systems of data repositories, deterministic agents, and retrieval-augmented models. These setups are designed to decouple growth from operational complexity.

3/ Building Sustainable (Data) Moats

When models and APIs are broadly available, defensibility tends to come from two places: proprietary data and tight operational integration into customer workflows.

On the data side, we’ve seen companies take different approaches:

1. Some fine-tune models on customer-specific data, e.g., a real estate brokerage might collect location-level sales data from existing clients to improve the prediction quality of future site selection models.

2. Others gain an edge by understanding structural document patterns across the delivery stack, e.g., an AI credit analyst who can reliably parse and standardize financials across formats, jurisdictions, and edge cases.

3. A third group builds semi-structured internal knowledge bases that accumulate over time, curated, expanded, and corrected by humans-in-the-loop. These repositories power the agents and create compounding product quality.

On the integration side, defensibility comes from becoming deeply embedded in the customer’s operation. If the system is closely tied to core workflows (i.e. decision-making, compliance, reporting), it becomes harder to rip out. When agents are wired into core processes — like claims adjudication, compliance opinions, or alert triage — they stop being tools and start becoming infrastructure. Replacing them isn’t just a software switch; it requires changing how work gets done.

4/ Building and Articulating Value Propositions Beyond Lower Costs

AI-first service companies often replace existing vendors and tap into budgets that are already being spent. In those cases, price alone rarely wins. What matters is whether the internal ROI from AI-driven delivery translates into better outcomes for the client, measured on the KPIs they already track.

Tenex, for example, competes with legacy managed detection & response providers not just on cost, but on precision and speed, promising higher true positive rates, fewer false positives, and faster detection and response times. In insurance, TPAs are expected to show improvements in loss adjustment expense, lower error rates, and better loss ratios, while also offering more transparency into how decisions are made. Legal tech firms like Covenant position on similar lines: faster turnaround, clear communication, and a fraction of the traditional cost.

These external KPIs are often mirrored internally by a north star metric focused on delivery efficiency, most commonly some form of revenue per employee. It gives an early signal of whether the automation and infrastructure are actually compounding, or whether the business is still scaling like a traditional service firm.

Software vs. AI-First Services: Pick Your Battle

As we are consolidating our findings on this space, we started debating the question of what’s the best way to bring this new technology to market. Through software, or through full-stack services?

The case for full-stack AI services is getting stronger. These businesses expand the market size in two ways. They make previously unattractive verticals viable by absorbing service complexity. And they go after larger budgets by replacing existing service vendors. Their entire organizational structure is built to maximize the ROI of AI implementation. When the offering is differentiated on critical KPIs, there’s often no real PMF risk because the budget already exists. Go-to-market can also be faster. Especially in industries with platform fatigue or slow adoption cycles, positioning as a service provider makes it easier to slot into procurement processes. That said, service businesses come with real tradeoffs. Operational complexity increases quickly. Margins depend heavily on the pace of automation. And scaling the business often requires orchestrating tech, people, and workflows in parallel.

Software remains the cleaner model. Gross margins sit in the 80 to 90 percent range. The cost to serve one more customer is nearly zero. Risk stays with the buyer. And when software is deeply integrated, defensibility compounds through usage, data, and switching costs. But the challenges are real here, too. Customers need to be convinced that black-box AI tools will actually deliver. Many software companies still face long sales cycles because trust and integration need to be earned before a deal is signed. In practice, many buyers still expect to purchase outcomes, not just tools.

We don’t fully agree internally. Louis leans towards software as a more efficient distribution vector. I find myself more drawn to models that start out service-heavy and move gradually toward software margins (maybe Louis is just a boomer, or I am just a GenZ cliche). It’s still early, and we expect the lines between software and services to blur even further. Neither model is perfect. The hard part is knowing which imperfections you’re willing to build around. Pick your battle.

A few weeks ago, I started using ChatGPT as a to-do list app. I just dump everything that’s on my mind into it, very “raw”, usually in voice mode. ChatGPT organizes these inputs into tasks, removes completed ones, and suggests next steps. When I want to get an overview, I simply ask, and ChatGPT returns an organized list of recently finished tasks and pending tasks grouped by priority.

Here’s an example for an input …

… and how ChatGPT organizes the list:

It’s not a sophisticated setup, not even a Custom GPT. It’s just a chat in a project that I’ve pinned to the top and called “Tasks”. I just ramble every open loop in my head, usually on my way to work, and let the model sort it out.

The only customizations are an Attio export to help with name recognition and spelling and some custom instructions, but to be honest, it worked surprisingly well even before I added those. Well, maybe not surprising in early 2025 — context windows have become so long, and LLMs have improved so much in the last years — but I think my 2022 self would be surprised that a general-purpose AI could outperform dedicated to-do list apps.

Why traditional to-do list apps never worked for me

This is a bit embarrassing for someone who’s a productivity geek otherwise, but I’ve never been a great user of traditional tasks managers. I’ve tried many — it’s a busy category with a lot of great products — but none of them stuck.

One problem is that it’s hard (at least for me) to truly make the to-do list app a comprehensive “system of record” style list that includes all of my tasks. Tasks come from everywhere — email, Slack, WhatsApp, meetings. If you’re not super disciplined in consolidating everything into one place, your to-do-list app doesn’t reflect the full picture, and you end up with multiple half-complete lists.

In theory, it should only take a few clicks to copy & paste a task over, but my guess is that except for a small-ish group of GTD heroes, most people aren’t disciplined enough to do this consistently. Another option is to set up automations (e.g. using Zapier) that automatically add tasks from everywhere to your to-do-list, or forward tasks via email to a special address that then adds the task to the list. Definitely possible, but again, not so practical for most people.

Then there’s the second problem: I kept forgetting to use them (don’t laugh, I know this makes me sound stupid). And if you can’t rely on yourself checking the to-do-list app, then for an important task you send yourself an email or put a post-it on your screen, which obviously defeats the purpose.

Why ChatGPT works better (for me, for now)

Using ChatGPT as my task manager works well because:

1. Capture friction drops to zero. Natural language is the best UI for this type of data entry, and voice recognition is now so good that it works perfectly in voice mode.

2. My tasks live where my attention already is. I use ChatGPT all day, so here’s no chance I’ll “forget” about it.

3. The model adds (some) structure/enrichment. This will get much better with more context and integrations, but even without that, it automatically groups subtasks, suggests next steps, etc.

So far, I haven’t had problems with context drifts or hallucinations. I think after a few weeks of heavy back-and-forth, the model might lose track of what’s done, so I might have to start a new chat and copy & paste the latest status over at some point.

It’s still a bit too early to tell if I will ultimately stick to ChatGPT as my to-do-list app (I’m still in the honeymoon phase, and I’ve honeymooned with other to-do-list apps before that didn’t become lasting happy marriages). But so far so good.

Where this is going

Right now, the system is still extremely dumb compared to how it will work in 1–2 years from now (trust me on that!). ChatGPT is not yet connected to my email, WhatsApp, calendar, meeting transcriptions, Slack, files, and other systems. Connecting ChatGPT with these tools will be the real game changer because then:

• It will not only track tasks, it will (at least partially) start doing them (e.g. drafting emails).

• It will know when a task has been completed so it will require even less input from me for the monitoring part.

• It will add/suggest tasks e.g. from meeting transcripts, Slack messages, and other sources.

• It will remind/nudge me, whether it’s with a built-in reminder feature (recently introduced by OpenAI) or by putting something into my calendar.

• With deeper contextual awareness, it will know which tasks are more urgent than others.

This is all very much possible with today’s AI, I just need ChatGPT to finally plug into my stack. ;-)

In the future, ChatGPT might also render a suitable graphical UI on the fly (e.g., displaying checkboxes that I can tick off by clicking). As much as natural language is excellent for task input, it’s not necessarily the best interface for every type of action.

What does this mean beyond task management?

Looking just a bit further ahead, with richer context from various data sources and greater intelligence, AI will handle more and more tasks autonomously. In many cases, users will only need to approve actions drafted or suggested by the AI.

If (and that is still an IF) AI kills traditional task managers, what does that mean for other software categories? Before you cry out “SaaS is dead” too loudly, keep in mind that task managers are very simple applications. Replacing CRMs or ERPs — with their intricate business logic, complex data structures, collaborative workflows, and permission models — is a different feat. I don’t think that AI won’t replace these within the next few years, but longer term, it’s a very real possibility (as Satya Nadella said a few months ago). And even if AI doesn’t completely replace the application, what if it replaces the UI layer? That creates big challenges, risks and questions for application-level companies, particularly those where the UI is key to their differentiation or stickiness.

As per our previous posts, vertical and highly specialized applications should have less to worry about in this scenario. Then again, there’s always the question of how fast the big players will move. Granola, Lovable, Cursor all emerged quickly in areas one would think should be dominated by large incumbents. Will the same happen in the personal productivity space and other horizontal software categories?

1. Code Generation is THE B2B Killer App of Generative AI

It’s far from the only one, of course. Generative AI is impacting software across many verticals, and the ultimate gen AI killer app is ChatGPT. But in B2B, and for a relatively homogeneous use case (compared to ChatGPT, which, like Web search, is used for everything), code generation stands out.

Since writing that a week ago, I’ve come across some real data that prove it. ;-) If we take these (slightly outdated) UBS numbers and break it down into the major categories, we get this list:

As you can see, code generation represents more than 28% of what UBS includes under “AI Native Revenue”. If we combine image, video, and audio generation into media/creative, that category tops the list, accounting for more than 38% of the total. I’ve found it interesting to see that taken together, code generation and media/creative account for more than ⅔ of the total. As a caveat, this is based on data from only 21 companies, but I doubt that looking at the long-tail would radically change the picture.

2. It won’t take too long until most code will be written by AI

One weekend ago, I built a level editor in Unity for a game my son is working on. (He only lets me touch the plumbing, I’m not good enough for the really interesting parts 😄). Most of the C# code was AI-generated. I barely understood half of it, but it works.

That raises a big question that I’m sure many people have asked themselves lately: what happens if we increasingly rely on AI-generated code? Are we comfortable pushing code to production that no human being has reviewed or understood?

You could argue that this isn’t new. For C# to run on a computer, it must be translated into CPU instructions (machine code) via several intermediate steps. I remember back in the Commodore Amiga days, games (and what was called “demos”) were coded in pure assembler, which sounds (and I think was) crazy. We’ve since moved up the abstraction ladder, trading raw opcodes for C, Python or JavaScript. Maybe in a few years people will find it laughable to use a language like this because we’ll be talking to computers in plain English?

So abstraction isn’t new. Developers have always relied on compilers to generate lower-level code that no human ever reads. What’s different is that in the past, those compilers were deterministic and rule-based. What’s new is the probabilistic, opaque nature of LLMs.

For what it’s worth, I strongly believe that we’ll get there and that it won’t take decades. AI-generated code will earn our trust quickly — because AI will simply outperform human developers. Once AI codes better than almost any human, the dynamic will be similar to self-driving cars. Will autonomous vehicles make zero mistakes? No. But if they prevent 90% of accidents, we should get them ASAP.

3. Cursor, Midjourney, ElevenLabs: PLG on AI Steroids

Pre-PLG, a software company’s growth rate was usually constrained by the speed at which it managed to hire and train salespeople. Companies like Slack, Dropbox, and Zendesk showed that product-led growth can enable faster scaling. Now companies like Cursor, MidJourney, and ElevenLabs are taking this to the next level … what I like to call PLG on AI steroids.

None of these companies achieved their growth by building out a massive sales team. You can’t sell that fast. It has to be bought. And the speed of adoption is driven by the insane value that users get, enabled by AI. Traditional PLG usually came with a 30 day trial for setup, education, feature discovery, etc. Some of these new AI tools compress the entire onboarding and activation process into minutes.

In 2012, I wrote about how SaaS companies should make the learning curve as smooth as possible and give the user as much gratification along the way as possible. Just like game designers need to teach the game to new users in many small steps, meticulously making sure that it never gets too difficult nor boring, B2B software companies need to create a frictionless and rewarding onboarding experience.

With AI, it’s as if Mario has discovered steroids (or rocket fuel). The jumps are bigger. The payoffs come faster. And the wow moments hit instantly.

The question is: what other categories will see this dynamic emerge? Who else can turn foundational models into a product with such extreme pull?

4. Where Does the Value Accrue in Code Generation?

There’s been a lot of debate about where value lands in the AI stack. Will applications like Cursor win, or will the value flow down to the model providers (in addition to Nvidia)?

The bear case for Cursor, which started out as a rather thin UX layer, was that most of their revenue goes straight to Anthropic. That view seems outdated, given that Cursor is now training its own models, purpose-built for code. As Michael Truell said in the interview with Lenny: “We definitely didn’t expect to be doing any of our own model development. And at this point, every magic moment in Cursor involves a custom model in some way.”

Poolside, a P9 portfolio company, started there from day one — training an LLM that is entirely oriented towards software development and that improves by completing millions of tasks across tens of thousands of real world software projects (what Poolside calls Reinforcement Learning from Code Execution Feedback). Their foundational model, which can be fine-tuned on how customers write software, what libraries and APIs they use, etc. is designed to enable developers to produce the best code they never wrote.

So while Cursor and Poolside took very different paths — Cursor initially focusing on UX and generating momentum with PLG, Poolside focused on R&D and enterprise readiness — it seems that they are converging on the same view: if you want win in code generation, you have to deliver the best possible user experience and be able to make money even if margins decrease — and for that, you have to own a large part of the stack, including the model.

I’m generally not convinced by the “thin wrapper” narrative that dismisses apps as shallow UI layers on top of someone else’s model. Over the past few years, we’ve invested in multiple companies that solve real customer problems and build meaningful products without training their own models. But code generation may just be a special animal.

Below, we share some key insights and a market map.

A Framework for Categorizing AI-first Companies

Before diving into the specifics of AI-first service businesses, it’s helpful to step back and consider the broader landscape of AI-first companies. After carefully reviewing hundreds of businesses, we’ve observed that these businesses can be grouped into three distinct categories, each with unique characteristics and implications for scalability, differentiation, and value capture.

1. Agentic Software (Autopilot): Fully autonomous AI systems capable of managing workflows from start to finish without human intervention, such as ScienceMachine (AI data analyst for drug discovery), Sphinx (agents for KYC & AML tasks), or Conveo (AI-moderated voice & video interviews). These solutions excel in highly structured and repeatable processes such as automated data entry or monitoring. While they promise SaaS-like or superior margins (they don’t have any “humans in the loop”), they can face significant barriers to adoption, as customers often demand extensive pilots to build trust and validate reliability before fully outsourcing a process to an AI.

2. AI-First Workflow Automation (Copilot): Tools that augment, rather than replace, human workers by integrating seamlessly into existing workflows, e.g., Harvey (legal AI-copilot), Ascend (automation of KYC/KYB workflows), or Console (IT support co-pilot). These solutions are ideal for complex tasks where human judgment is essential, but automation can benefit portions of the process. With their familiar SaaS pricing models, such as subscriptions or usage-based fees, these have similar scalability levels and business models as the previous generation of SaaS software.

3. AI-First Service Businesses (Service Model): These companies aim to disrupt service markets, often presenting themselves as traditional service providers but natively adopting AI to streamline labor-intensive processes while increasing their margins. By focusing on services rather than software, they can capture more substantial portions of customer budgets (e.g., recruitment costs or compliance spending). While they may not achieve the high margins of pure SaaS models, their hybrid strategies enable them to penetrate industries that have historically resisted technological disruption.

Let’s now focus on the last category.

The Unique Value Proposition of AI-First Service Businesses

We won’t go as deep as Louis’ post from last year here. In short, we believe these businesses are interesting for two main reasons.

One, unlike traditional SaaS businesses that go after IT budgets, AI-first service businesses go after labor/people budgets. They deliver complete operational outcomes, capturing a share of spending that can be substantially larger — potentially encompassing entire operational budgets rather than just an IT line item. Imagine a typical bookkeeping scenario: while tools like QuickBooks cost $50–$150 monthly, the bulk of the expense — $500–$2,000 — goes to labor, i.e., the accountant. Companies like Pilot flip this equation: they provide an end-to-end bookkeeping service, combining AI automation with human oversight. They charge $349 for pre-revenue startups up to $999+ per month for growing companies, automating core tasks, and capturing a significantly larger share of the total spend.

Two, beyond increasing ACVs by delivering operational outcomes, AI-native services could allow startups to turn historically low-margin or labor-intensive service verticals into high-margin opportunities for software-first startups. Exemplarily, traditional European property management businesses, such as Vonovia or Deutsche Wohnen, rely heavily on manual processes and fragmented software. They manage routine tasks like tenant communication, maintenance scheduling, utility billing, rent collection, and administrative duties through labor-intensive efforts, resulting in high costs and limited scalability. In contrast, AI-first startups like Buena, Ralph, Hallo Theo and Arbio operate as genuine service companies, blending advanced automation with human oversight and regular in-person support.

These businesses automate tasks such as tenant requests, rent collection, contract processing, maintenance coordination, and financial management. Buena leverages automation to streamline new leases, tenant care, utility billing, and accounting, significantly cutting administrative burdens. Ralph enhances the client experience through automated customer service, real-time document management, and efficient handling of maintenance requests and property inspections, all backed by a dedicated contact person — see a screenshot of their FAQ below.

By automating these manual processes while maintaining human involvement, AI-first companies significantly reduce operational costs, increase service efficiency, and transform traditionally low-margin property management into high-margin, scalable business opportunities suitable for venture investment.

Our Market Map

After mapping 500 businesses during our research, several industries stood out as prime candidates for disruption by AI-first service businesses. These verticals share common traits: high labor intensity, structured workflows, and mounting pressures to reduce costs, improve efficiency, or deliver outcomes faster.

• Legal Services: AI can automate time-consuming tasks like drafting contracts and reviewing evidence during the discovery phase, where documents, emails, and records are collected. This frees lawyers to focus on strategic work. Human oversight ensures reliability in complex or high-stakes cases. Legal Process Outsourcing (LPO) has existed for years, but AI now enhances it by increasing speed and accuracy. EvenUp in personal injury cases and Atticus in employment and disability law are two interesting examples.

• Financial Services: Processes such as bookkeeping, due diligence, and compliance are well-suited for AI automation as they are repetitive and rule-based. Automating these tasks reduces errors, saves time, and enhances accuracy. Companies like Pilot illustrate this approach by providing SMEs and startups with comprehensive AI-driven financial services, integrating bookkeeping, tax preparation, CFO support, and R&D tax credit management into one cohesive back-office platform.

• Human Resources: AI can streamline repetitive tasks like candidate screening, scheduling interviews, and matching talent to job openings. This frees HR teams to focus on fostering workplace culture and addressing strategic initiatives. Mercor, which quickly scaled to >$70M in ARR, is a prime example in this category.

• Logistics: In logistics, AI is transforming areas such as claims processing for lost or damaged shipments (Claimit), automating B2B invoice recovery (Respaid), and even building AI-driven logistics carriers (Cartage) that manage entire shipping processes from start to finish.

• Insurance: In insurance claim processing, companies like ClaimSorted, Reserv, and Strala use AI-driven Third-Party Administrator (TPA) solutions (TPAs manage insurance claims processing on behalf of insurers) to transform a traditionally labor-intensive, low-margin industry into a scalable market. AI automates processes such as First-Notice-of-Loss, claims routing, and triaging, significantly reducing manual workloads. Real-time analytics accelerate fraud detection and reduce risk, while end-to-end claims visibility enhances insurer decision-making.

• IT Services: AI is revolutionizing IT services by automating complex tasks such as legacy system modernization, custom software integration, and cybersecurity threat detection. This transformation enables IT professionals to focus on strategic initiatives, enhancing overall efficiency and service quality. For instance, Mechanical Orchard specializes in modernizing critical legacy applications without disrupting their current operations.

• Compliance and Cybersecurity: Companies like Oneleet, Bastion, and Delve leverage AI to automate compliance and cybersecurity tasks, including threat detection, continuous monitoring, vulnerability assessments, automated security audits, and regulatory adherence.

• Property Management: Lastly, as discussed above, AI-first property management companies like Buena, Ralph and Hallo Theo automate key workflows while maintaining human oversight.

All of the above are just observations we’ve gotten mapping and speaking to companies in this space, which we find incredibly exciting.

If you want to be added to the map, chat to us about this nascent field, and help us develop this thesis, please write us here!

One of the concerns we’ve heard multiple times about B2B marketplaces is that they are tech-enabled service businesses (or digital traders) rather than software businesses. The line between these two archetypes of business models is somewhat blurry, but the argument usually revolves around 4 points:

1. Sales: Salespeople employed at a B2B marketplace do not only work on acquiring new clients, be they buyers or sellers. They are needed to handle transactions between existing buyers and sellers and, as such, impact B2B marketplaces’ gross margins.

2. Support: Human support is required for ongoing transactions and needs to scale with the number of transactions. This also impacts B2B marketplaces’ gross margin.

3. Buyer/seller insurance: Claims between dissatisfied buyers and sellers also require human support and must be handled by the employees of B2B marketplaces, which are very often merchants of records (i.e., part of the transaction).

4. Finance: digitizing money flows in a B2B context is often more complex than plugging a payment API and, therefore, involves managing large transaction values, working capital, AR, and AP workflows. All of these can be people-intensive and difficult to scale.

Every marketplace is different. Depending on the nature of buyers, sellers, and transaction flows, some of the processes above can be more or less difficult to automate with traditional software.

Now, as we’re all spending time with companies building AI agents, we also noticed that each of the items above: sales, support, insurance/claim management, and finance workflows are prime examples of processes that AI agents have a chance to automate in a not-so-distant future. We’ve seen many examples lately of sales, support, claim management, and finance-focused AI agent businesses; what if they were implemented in a B2B marketplace context?

The argument is also that B2B marketplaces (or B2B tech-enabled services) are in a great spot to launch these agents because they have the data, the customer relationships, and they are software-first organizations compared to their traditional competitors.

About a year ago, we wrote about AI-first service businesses here. We’re increasingly curious about what such an AI-first approach could look like in a B2B marketplace context. These agents could finally provide an elegant answer to the scalability (or gross margin) challenge that many marketplaces face. In a world where the defensibility of traditional software businesses is being questioned, these B2B marketplaces often play in vertical markets with much less competition but sometimes display strong potential for network effects.

At a more macro level, we remain convinced that there are still many industries where trading flows between industry players will be digitized over the next few years. This digitization process could happen in a centralized fashion through new B2B marketplaces or more decentralized by deploying software (and agents) in traditional players.

We’re excited to explore these opportunities with our existing portfolio companies and are looking for new ones to add. If you’re building or working on that, please write us here. We’d love to chat!

At Point Nine, we’ve always believed in the power of community and throughout the years we’ve been building a tight-knit community of entrepreneurs building predominantly in B2B globally that care about giving back and supporting each other’s on their journeys. Through the years, we’ve organically brought founders together online and offline and developed an events calendar that caters to their needs.

We’re excited to have Laura join us to bring what we’ve done over the years to the next level.

Laura is no stranger to early-stage startups nor to Point Nine. After starting her career in brand strategy she joined our portfolio company cargo.one to build their marketing function and served as chief of staff working closely with the founders. Since then she’s advised startups on planning, OKRs, and how founders should develop their company OS.

As we thought about the right profile for our first-ever Director of Platform, we couldn’t help but think that someone who worked operationally inside an early-stage startup was key. An even stronger plus was that they’d been part of our community and knew the principles and values that guide it. In addition, the feedback we collected from portfolio founders that Laura was advising made it clear we had to bring her on board full-time.

The final bullet came directly from Oliver, one of the founders of cargo.one, who told my partner Louis when we did references that she was “overqualified for the role” :)

We couldn’t be more excited to welcome Laura to our small but mighty team. Our equal partnership now can count on Laura to support our founders in achieving their goals, driving performance, and beating the odds.

On behalf of everyone at Point Nine — welcome Laura

With so much up in the air, I wanted to share a few thoughts as we get ready for another wild year in AI. Don’t expect lots of bold predictions (my crystal ball is as blurry as ever), but here’s where my head’s at as 2025 begins.

1) Pre-training might approach diminishing returns, but it’s too early to declare the “end of scaling”

There’s a growing sentiment in the industry that we’re approaching “the end of the scaling laws”. The view is driven by the fact that GPT-5 hasn’t been released yet and that most of the (extremely impressive) recent improvements to OpenAI’s products stem from other innovations. As you may have seen, Ilya Sutskever, co-founder of OpenAI and SSI, recently added fuel to the fire by declaring the end of the pre-training era.

“The end of pre-training”,“the end of the scaling laws” and “the end of scaling” can mean different things, so it’s worth clarifying what exactly we’re talking about. The scaling laws for LLMs, described in a landmark 2020 paper by Jared Kaplan and several OpenAI researchers, state that model performance improves with larger models, more training data, and more compute. The optimal balance between model size and dataset size for a given compute budget was further detailed in the famous Chinchilla paper in 2022. Both papers state that each incremental increase of any of the three variables yields a smaller improvement than the previous one. So if we observe diminishing returns, it doesn’t make sense to talk about the end of the scaling laws: quite the opposite, this is exactly what the scaling laws predict.

Maybe this is just semantics, and perhaps what people mean when they talk about the “end of the scaling laws” is that we’ve reached a point at which scaling models further doesn’t yield practically meaningful returns. Industry experts have different opinions on the topic, and I guess no one really knows, but here are a few things to keep in mind.

First, performance improvements have never been solely about scaling the pre-trained model. Adding more parameters, data, and compute has been a key driver of the huge improvements from GPT-2 to GPT-3 and from GPT-3 to GPT-4, but it wasn’t just brute force. Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF) were critical in making the models useful and played a key role in making ChatGPT so good. (1) The same is true for the new o1 and o3 models, where the key innovation was to force the model to “think” before it answers, breaking down a bigger problem into smaller, more manageable steps. (2)

Second, while general-purpose foundational models have already been trained on most of the text on the Internet (a key argument of the people from the “end of pre-training” camp), specialized domains like biology or chemistry remain underexploited. So there’s still huge potential for progress by training on more domain-specific data. It’s an open question to what extent this will lead to performance improvements outside of the specific domain (but there’s evidence that it works for code, i.e. if you train an LLM on more computer code, it will get better at general reasoning). Similarly, there are high hopes that data in different modalities, especially video, and synthetic data will solve the data saturation issue, but experts disagree on the extent to which this is going to work. (Synthetic data definitely works for coding; in other domains, it’s less proven)

Finally, even if we’re approaching a point where scaling pre-training becomes prohibitively expensive, we’re only starting to find out how much better models can get with greater inference-time compute. O1 has shown that you get much better answers if you give the model more time to “work” on a problem. With more compute, models can think through more steps and increase the likelihood of reaching the right answer further. (3)

Net net — my best guess is that LLMs will continue to get better with more parameters/data/compute, but the improvement curve won’t be as steep as it was in the past, and more and more attention will go to everything that happens post pre-training.

2) There won’t be another AI winter …

When a new technology gets hyped up, inflated expectations are often followed by a deep trough of disillusionment, which is why many people worry that the current phase of excitement will lead to an(other) AI winter.

I don’t think so.

There will, of course, be lots of failures. Pilots that don’t convert. Startups that go bust (including many that have raised 10s of millions before strong PMF). Disillusionment in areas where products fail to meet expectations or perform as advertised. And probably some spectacular failures of companies that have spent hundreds of millions on training models and didn’t manage to turn the investment into differentiated products with a sustainable competitive advantage.

But there won’t be a big across-the-board AI winter where people question the value of the entire field. AI already delivers way too much value today, be it in coding, medical transcriptions, translations, customer support, or as a productivity booster for tens of millions of people. I also think that the amount of capital and talent flowing into AI in recent years will ensure continued progress at a high pace, even if pre-training isn’t the main driver anymore.

So if there is another AI winter, it will be a very mild, Californian winter, as Richard Socher recently said on a podcast. No Berlin winter.

3) … but some of the high-flyers won’t make it

In the last few years, many AI startups grew very quickly from 0 to a few million dollars in ARR (and some to much more), at a pace that was extremely rare in the past. Several factors contributed to this phenomenon:

• It has become much easier to build AI-powered products, with surprising, impressive new capabilities that wow users and buyers. In some domains, AI has surpassed a quality threshold, unlocking massive demand and allowing many players to gain momentum, even with similar products (e.g., writing assistants).”

• In the wake of the ChatGPT launch, AI has become a door opener. Every company wants to try AI tools and solutions. Getting companies to do a pilot has become much easier. An example is legal tech. It used to be a laggard industry in terms of tech adoption; there’s been talk about AI for many years, but not much has happened. ChatGPT has catapulted the topic to the top of the attention of every major law firm. According to Clio’s latest Legal Trends Report, AI adoption in law firms has skyrocketed from 19% to 79% in just one year.

In many cases, startups deliver tangible value to customers in ways previously unimaginable. However, I’m afraid that many fast-growing startups will plateau when churn kicks in and pilots don’t convert. This risk is especially pronounced for easily replaceable point solutions (simple to adopt but just as easy to switch away from), add-on tools (temporarily successful but unsustainable if incumbents integrate similar AI capabilities quickly), or human-in-the-loop products, where revenue traction may not reliably indicate PMF. (4)

The AI wave is lifting many boats, but not all of them will stay afloat. This is, of course, typical for big technology waves, so it’s not a new phenomenon.

4) Startups will solve the “last mile problem” of AI

When ChatGPT arrived, many people in the tech ecosystem (myself included) asked themselves: If AI keeps getting better at this pace, what’s left for startups to build? Won’t OpenAI, Anthropic, or Google’s latest LLMs eventually do everything? Do you still need specialized business applications if in a few years you have an extremely intelligent AI system that has access to all of a company’s data?

These are valid concerns, but based on what I’ve seen in the last two years, I think it has become increasingly likely that in spite of (or, maybe paradoxically, because of) the fast increasing capabilities of foundational models, there will be more, not less, opportunities for AI startups. The idea is that the more capable the models get and the more people try them, the more startups are needed to solve the “last mile” problems that models alone can’t address.

There are a few reasons why better models could expand the opportunity set for startups.

A) Rapidly increasing expectations

When models could barely generate coherent text, a good enough summary or response was impressive. Now that GPT-4, Gemini 2, and other models can write essays, debug code, and much more, our expectations have shifted. Businesses want AI solutions that are reliable (no hallucinations), accurate (fact-based and grounded in company data), and trustworthy (secure and explainable).

B) Integration is hard

Enterprises must integrate models into complex systems, ingesting data from a variety of sources and in different formats, integrated with custom workflows, and ensuring outputs meet domain-specific requirements. RAG (retrieval-augmented generation) sounds simple in theory, but in practice, you’ll have to overcome various challenges. How do you chunk, store, and rank enterprise documents effectively? How do you manage latency when retrieving and feeding data? How do you prevent irrelevant or misleading context?

C) Agentic systems further increase the surface area

There’s little doubt that the future belongs to AI tools that can autonomously complete multi-step tasks. But if you give so much power to an AI, making sure that the system runs safely and reliably becomes exponentially more difficult and important.

If foundational models expand the opportunity surface faster than they can provide the complete solution covering the last mile, we’re entering a golden age for AI startups that take raw capabilities and turn them into robust, enterprise-ready products, so let’s hope that the theory proves right. 🙂

5) “Virtual employees” might turn out to be a gimmick

12–18 months ago, a fascinating new type of AI startups emerged: companies that offer digital workers with human-like attributes (and sometimes faces and names) to automate end-to-end jobs, e.g. in sales and customer service. If you’ve been to SF recently, you’ve probably seen Artisan’s billboards all over the city.

It’s a super innovative and fresh idea. Since their digital workers usually use the same tools as existing, human employees, these startups piggyback on existing platforms and minimize integration effort. It’s also an opportunity to attack incumbents with differentiated packaging and a potentially disruptive pricing model. For customers, it’s a very compelling value proposition: Keep your existing software and workflows, just add some AI employees to take over some of the work at a lower price.

So there’s a lot to like. However, I wonder if having “AI employees” with human-like attributes really makes sense in the long run, or if this is a temporary hack that allows startups to quickly gain traction in the current phase of AI adoption. I’m leaning towards the latter. A lot of jobs might have to be reconfigured if AI turns out to be good at some parts and less at other parts of a human’s job. For example, if AI can handle 80% of an SDR’s tasks but only 25% of an AE’s, you can’t just replace all of your SDRs with AI SDRs. There’s still a lot to be figured out as we adapt to working with intelligent software and agents, but my hunch is that AI employees with faces and names won’t be part of the endgame.

6) With agentic AI, we‘ll all have to rethink human-computer interaction.

With agentic AI — models that can browse the web, execute code, use external tools, or handle transactions — we have to rethink human-computer interaction from the ground up. We’re not used to giving software so much power, and one of the key challenges will be defining the boundaries of what these systems can and cannot do independently.

Imagine having an AI agent for travel bookings. Even a seemingly simple task like booking a flight can’t be easily delegated to an AI agent, as it requires trade-off decisions, such as choosing between a faster connection or a lower price. Even if your AI agent knows your general preferences, there’s a high probability that it won’t get it right every single time in every specific situation. Now think about giving an AI agent the keys to autonomously deal with complex, multi-step workflows (and to interact with other AI agents!) in a business, where the stakes are much higher.

Lots of challenges must be addressed as companies allow agentic systems to handle more and more complex tasks with less and less human supervision. A useful analogy might be training and managing a coworker who gradually gains higher levels of permission as they demonstrate competence. However, as noted earlier, such analogies might be akin to the skeuomorphic UI of the early iPhone (temporarily helpful but quickly outgrown).

I’ve spent most of the past 25 years, first as a founder and then as an investor, focused on building and backing web applications and software designed to improve human-computer interaction. By tomorrow’s standards, much of that software was pretty dumb. (5) The emergence of intelligent agents requires entirely new UI paradigms and I’m super excited to see how the smartest founders will define the future of human-computer interaction!

(1) I was going to write “smart” but that would invoke the inevitable response that these models aren’t smart but just stochastic parrots that are excellent at appearing to be smart. Don’t feed the trolls. ;-)

(2) ICYMI, OpenAI has taught the O1 model to use “chain of thought” before answering, which used to be a highly effective prompting technique before O1 came out.

(3) So Nvidia shareholders have a good hedge. If companies use less compute to pre-train models going forward, there’s a good chance that they will use more at inference-time … which could be much more, because in this case the need for chips grows with the number of users.

(4) If you want to build, say, an AI accounting product and you start by selling an accounting service with some automation and a lot of humans in the loop, you’re not proving much because there’s a clear, existing market for accounting services. The real test is if you can over time remove the humans in the loop. Doesn’t mean that starting with humans in the loop can’t be a great strategy, it just means that the typical steps of how startups prove PMF are reversed.

(5) Fun fact: my first internet startup, a comparison shopping engine I founded in 1997, used agents to retrieve pricing and shipping cost information from online shops (but those agents weren’t intelligent).

Germany has always had a complicated relationship with share-based employee incentivization. For lack of (easily feasible) alternatives most start-ups have resorted to contract-based virtual schemes which merely simulate a traditional equity-based incentive plan(1). Such a scheme is easy to implement and avoids taxation before there are actual proceeds, but comes at a huge cost: The proceeds are taxed at the full income tax rate (currently up to about 47.5%).

The problem

The obstacles faced by German companies trying to achieve the privileged taxation awarded to shares are manifold and substantial:

1. The beneficiary (employee) must receive real securities to “lock in” a base line above which (favourable) capital gains tax (CGT) applies. German tax law does not recognise the “derivative” most used in other countries for this purpose, the stock option, as a tax-relevant event before it is exercised.

2. Granting shares triggers an immediate (income) tax liability unless the beneficiary pays a purchase price at fair market value (FMV). Structures around this, e.g. a negative liquidation preference, are complex to implement and at best suitable for a few high-level recipients.

3. Transferring shares (instead of an option) in a German GmbH (the most common company format for SMEs in Germany) is tedious, requiring notarization and registration in the commercial register.

4. Share ownership in a German GmbH comes with indispensable legal rights even for micro-shareholders who otherwise have no connection to the company, in particular far-reaching information rights (sec. 51a GmbHG).

5. According to German corporate law some important shareholder resolutions can only be implemented promptly if they were taken unanimously — including the consent of potentially untraceable ex-employees.

Attempts to fix it

The German government tried to fix some of these issues in 2021 but the result, laid down in a new section 19a of the German income tax law (EStG), wasn’t well received. The new rule foresaw the deferral of the tax liability from the time of grant to the sale of the share (which was commendable), but only if the employee was still with the company at that point. This made the law a non-starter: If the employee left before an exit, the tax would become payable without there being any proceeds — a classic case of dry income.

With effect from January 2024 the German government sought to remedy this and certain other flaws with a comprehensive overhaul of sec. 19a EStG. These are the most important changes:

• The tax liability can now be deferred beyond the departure of the employee if the employer guarantees its payment(2).

• The expiry date of the deferral (which would trigger taxation even if the beneficiary is still employed with the company) has been extended from 12 to 15 years and can again be further prolonged if the employer guarantees the tax liability.

• The criteria for a company to qualify for sec. 19a EStG have been generously broadened to cover even “large” SMEs: ≤1.000 employees and ≤EUR 100m revenues or ≤EUR 86m assets. It’s not a problem if a company exceeds any of these criteria at the time of allocation if they were met in one of the preceding six years (before: one year).

• Securities don’t need to come from the company but may be “secondaries” transferred by an existing shareholder (although the issuer must still be the company).

• BUT: The circle of potential beneficiaries remains limited to employees, making the scheme unsuitable for companies with lots of advisers and freelancers(3).

The new rules have been well-received, though some challenges remain. Yes, the dry income issue (#1 and #2 in the earlier list) has now been sufficiently addressed, but ##3–5 have not: There’s still a need to transfer real securities and there are still the minimum information rights and governance constraints if these securities are shares. For a micro-GmbH with only a few — trusted — employees this might be manageable, but once there are more than a handful of employees the admin effort, governance challenges and nuisance potential of disgruntled ex-employees continue to make a share-based plan challenging.

There are attempts at workarounds, in particular the bundling of employee stock in a company-controlled holding entity (e.g. a limited partnership in the form of a GmbH & Co. KG). Such a structure helps with governance, but also means additional admin requirements and costs. And while it may be easier to transfer shares in a partnership than in a GmbH, changes on that level will need to be reflected in some shape or form at the GmbH level. A further entity to warehouse the shares can make this easier, but creates even more complexity. In light of all this a holding setup may be the right choice for companies with adequate resources and head-office functions, but rarely for early-stage start-ups.

A workable workaround?

Could the revised sec. 19a EStG end like its predecessor and be mostly ignored? Not necessarily. A startup in Berlin, Billie GmbH, recently managed to secure a so-called payroll tax ruling (“Lohnsteuer-Anrufungsauskunft”) which confirmed that its incentive plan could be based on profit participation rights (or, shorter, participation rights — “Genussrechte”) rather than real shares. Simply spoken, participation rights give their holder the right to share in distributions and exit proceeds as if she was holding real shares, but don’t come with the information and governance rights of real shares. In essence, a participation right is not all that different from the contractual claims granted by a typical virtual incentive program.

Participation rights have always been part of the group of securities considered eligible for capital gains tax rules. But because they are, compared to shares, a fuzzy concept, it’s not easy to say at what point the package meets the label (more on this below), and a ruling of the highest German tax court (Bundesfinanzhof) hasn’t exactly made things easier. On top, participation rights can help with the German corporate law challenges, but not — until recently — with the dry income problem if they are granted below FMV.

Things look different now: The ruling of the Berlin tax authority combined with sec. 19a EStG should give Billie sufficient legal certainty. Yet its scheme cannot simply be replicated by others because the ruling is not a law and only binds the Berlin tax office vis-à-vis the applicant (Billie). One can hope that it will not contradict itself when dealing with similar cases, but there’s no certainty, and the tax authorities in other German states may not even feel a “moral” obligation to take the same view.

German start-ups shouldn’t be deterred, however. Even if a particular incentive scheme based on participation rights ultimately doesn’t find the approval of the tax authorities, it would have to be treated like a traditional virtual plan. This would mean a higher tax rate at time of payout, which would be frustrating, but not worse than what would have been the case if the traditional approach had been pursued from the beginning.

So is it even worth applying for a tax ruling like Billie did? At this point we think yes, because even if the ruling takes a year to obtain it would provide legal certainty and, if positive, allow you to be much more assertive about the tax benefits of your scheme(4). And if it turns out negative you may still be able to adjust the terms so that at least for the future the plan meets the requirements. Depending on circumstances, even a more complex structure involving a separate holding entity (see above) might still be considered at this stage.

There are two things to keep in mind: First, as of today, it is unclear whether participation rights can be issued with a hurdle price(5). Hurdle prices are common in virtual programs for allocations made after the initial funding round(s) in order to limit the award (mostly) to the value created after the allocation (making them similar to the exercise price in case of stock options). Second, the Billie plan included a small consideration to be paid by the beneficiary based on the assumption that participation rights require a financing element of about 3–5% of FMV.

It may be worthwhile including both these things in an application for a tax ruling with the aim of allowing hurdle prices and getting rid of the financing element (or at least reducing it to the one Euro-minimum German law requires for shares), but keep in mind: The more you divert from what’s traditionally considered characteristic for a participation right, the longer the process may take and the greater the risk of a rejection.

What about existing virtual plans?

What does this mean for German companies which already have a virtual plan in place? Simply converting existing virtual shares (i.e. the potential future claim they represent) and deeming the resulting participation rights to have been issued already at the time when the virtual shares were allocated will not work. The next best solution, canceling the virtual plan entirely and replacing it with participation rights, seems straightforward, but may also encounter resistance by the tax authorities(6).

This leaves one option which should definitely be unproblematic: Terminating the virtual plan for the future, i.e. locking-in a potential payout at the current value of the company, and issuing participation rights instead. If these are issued without a hurdle price the company might be obligated to pay double up to the locked-in value, however this extra benefit for the beneficiary could be reflected, even if imperfectly, by reducing the number of participation rights she receives.

A look beyond the border

Where does all this place Germany in comparison to the major VC markets, particularly the UK and the United States? Assuming the “Genussrechte-model” indeed turns out to be a workable solution, the picture is quite favourable:

As the table shows, the UK (in case of EMIs) and US (for ISOs) get around the problem of dry income taxation at grant by treating options like real shares and requiring them to be issued with a strike price at (broadly) FMV. The exercise of the options is then effectively treated as a “non-event” and taxation only kicks in when the shares are sold.

Germany, in contrast, did not want to go as far as treating options like shares, but instead decided to allow the transfer of “real value” (= securities without a strike price) and defer the tax payment to a time when the value is realised. Assuming the Genussrechte-model is also feasible with hurdle prices, Germany’s new scheme would be more flexible than EMIs and ISOs in this respect.

Otherwise the picture is mixed: Germany’s CGT is higher than what beneficiaries in the UK and the US ultimately have to pay (currently generally 26.375% vs. 0 to 24% at most), but it does not come with the onerous exercise requirements in the UK and the US. In particular where a beneficiary leaves the company before an exit (which in today’s labor market is probably the rule rather than the exception), the need to exercise the options (and pay the exercise price) can become an insurmountable obstacle.

In terms of eligibility sec. 19a EStG is significantly more generous than EMIs in the UK, where SME criteria are tighter and the overall volume is capped. The US has no size limitations for the issuer, but ISOs are subject to a USD 100k annual cap per employee.

Time to change course?

It’s time for German startups to rethink the traditional “virtual shares default”. Even if employees may not (yet) insist on the implementation of something “better” for a while, opting for a 19a-compliant scheme as early as possible makes sense: Besides the marketing potential such a plan enables the company to achieve the same after-tax benefit for employees with a smaller number of securities than with virtual shares, meaning less dilution for shareholders. In addition the new rules are likely to become standard rather sooner than later, so the earlier you address this, the fewer beneficiaries might ask awkward questions in the future about virtual shares they got before the change. As of now, participation rights seem like the most suitable route forward, at least for smaller start-ups and despite the questions around consideration needs and hurdle prices.

__________________________

Notes: 1) The basic idea of a virtual stock incentive plan (“VSIP”, often also referred to as “VSOP”) is that certain exit events are defined which trigger a (contractual) payment to the employee depending on the proceeds payable to common shareholders and the number of virtual shares owned by the employee. 2) If an employee leaves and must return unvested (or potentially even vested) securities or if the ultimate sale of (vested) securities happens below the FMV at time of grant, the tax payment obligation is triggered, but only on the proceeds (if any) actually received by the employee. 3) In addition, only German tax subjects, i.e. employees living in Germany, can benefit from the 19a-privilege. For employees who are tax-resident outside of Germany the rules of their country, potentially modified by a double-taxation treaty with Germany, count. As of today sec. 19a EStG also requires the employee to be employed at the issuing company, however this limitation is scheduled to be removed with effect from fiscal year 2024, so that also employees of subsidiaries may benefit. 4) In addition, the FMV of the benefit issued must anyway be declared to, and confirmed by, the tax authority each time a benefit is granted. 5) Sec. 19a talks of securities being offered below market value, which is a necessary requirement for a rule that deals with deferring taxes on the implied benefit. But sec. 19a not applying (if the hurdle is set at FMV) doesn’t necessarily mean the right is no longer a participation right. If it remains one, and therefore subject to CGT rules at time of sale, sec. 19a wouldn’t even be needed. 6) The tax authority might deem the new 19a-plan an early payout of the VSIP (not in cash, but in kind, and therefore still taxable). While this seems remote (no payment was due yet on the VSIP and a tax payment obligation would undermine the deferral of the 19a-plan), there is some residual risk. Another feature that could be included in a future tax ruling.

I just spent the last 3 weeks in the US and took this as an opportunity to meet as many people as possible to learn and reflect on the state of things in software in 2024. Here are some high-level learnings that might be worth sharing.

1/ Software engineering is changing (even faster than expected)

Between code generation foundation models like Poolside and new IDEs like Cursor.ai enabling multiple code generation models within the same UI, it’s now evident that software engineering is changing. I met many engineers who had tried early/ier versions of these last year and weren’t very convinced but now couldn’t do without them. Menlo Ventures just surveyed 600 IT decision-makers in the US and reported that 51% were already using code copilots (see chart below).

To that extent, code generation has reached what we called (back in 2018) the MAP (Minimum Algorithmic Performance) to get broad adoption. Anyone in software development today should use these, and all of us in software should likely bake the assumption of higher levels of productivity in software engineering in the way we build and invest in software products and businesses.

2/ (The) Software (industry) is changing

The consequence of the above is that the broader software landscape is changing. Smaller engineering teams can build more comprehensive software products faster and cheaper. Companies like Linear get to tens of millions in ARR with only tens of employees.

The positive consequence is that specific opportunities requiring considerable software development investments before getting to market will be easier to seize. The negative is that other markets will become increasingly competitive.

This is good news for vertical software companies with long product roadmaps, comparably little competition, but that need comprehensive software offerings (and higher ARPAs) to make the opportunities they go after large enough (more on that here). Beyond faster software development, the new possibilities enabled by LLMs (e.g., document generation, text extraction) or simply new API building blocks more broadly available (e.g., payment, lending APIs) will unlock new opportunities for significant value creation through software. All of the above (faster software developments, more building blocks available) compounds in a fascinating way.

3/ The software opportunity will likely expand

We might wonder what lower and lower software development costs mean for the broader software opportunity.

First, we are convinced that one of the bottlenecks remains the ability of software companies to imagine the right software features (i.e., do the right product management) iterating with customers. This takes time and won’t go away anytime soon.

Second, we are convinced that there are still a lot of digitization/automation opportunities that will be unlocked thanks to the above: digitizing vertical B2B markets or B2B commerce (i.e. B2B marketplaces), bringing software to the physical world through (new) sensors, building software-enabled hardware or building AI-first service businesses that look like software businesses, for example.

Last and obviously, the fundamentals of software companies remain as attractive: very low delivery cost, high margin, recurring cash flow, etc…

One way to put it is that AI might shrink existing market segments but will also make the overall software opportunity (pie) significantly larger. The slide below, published by Index, shows it simply and well.

4/ Use cases for software are broadening dramatically and attracting talented people

Building on the above, we’re noticing a dramatic expansion in the number of use cases AI allows software companies to tackle.

In parallel, the shared conviction that LLMs represent a technological disruption that’s sustainable and worth riding is attracting a lot of people to start businesses. It’s happening in Europe, but it’s happening in the US at a significantly faster pace. Without commenting on the economic and political situation, I saw many European founders moving to the US to start LLM-enabled businesses.

Here’s just a list of cool software ideas I came across in the past 3 weeks:

• Foundation models for data science

• Vertical foundation models (e.g. bio, material, …)

• Verticalised versions of information retrieval and data normalization across multiple databases for finance, healthcare, pharma…

• Verticalised generation of documents to automate the answer to RFPs in the public sector or construction,

• Fully automated interviews for recruitment,

• …just to name a few!

5/ An update on AI-first services businesses

Beyond “pure software” opportunity and, as we wrote in this blog published earlier this year, we (and many others) now believe that specific repetitive tasks humans perform will be partially or fully automated through LLMs.

People are attacking this opportunity by building AI-first service businesses from the ground up or buying, merging, and automating existing businesses.

I heard of multiple roll-ups in accounting, healthcare, and customer service on both coasts of the US (none in Europe). Another exciting thing I encountered was a prominent film production company raising money to train a foundation model on their film archives.

Other exciting ideas we came across lately were rebuilding market research firms with AI-first survey definitions and/or AI-led interviews, rebuilding IT consultancies, staffing agencies, or expert networks, all AI-first.

6/ Organisations are changing

Last but not least, organizations are and will be changing. Beyond the increased productivity of engineering teams mentioned in the first part, there’s now an increasingly strong belief that AI agents will perform specific tasks inside organisations such that they (and work) will change more broadly.

I’ll finish with this quote from Jensen Huang at Nvidia (here), who imagines a world where Nvidia remains an organisation with 60 direct reports for Jensen Huang, “only” 50,000 employees (which they more or less already have today) that will “all be CEOs” of 5M AI agents 🤯

Things are moving so fast these days!

Gladly co-founder and Point Nine Capital advisor Michael Wolfe is releasing a series of essays to help founders manage themselves and their startups. We’ll discuss founder motivation, productivity, mental health, relationships with co-founders, hiring and managing people, building a strong culture, and keeping people aligned and working on the right things. You can follow the full series at uninvent.substack.com.

“In effect, there are two different ways to run a company: founder mode and manager mode. Till now most people even in Silicon Valley have implicitly assumed that scaling a startup meant switching to manager mode. But we can infer the existence of another mode from the dismay of founders who’ve tried it, and the success of their attempts to escape from it.” — Paul Graham, Founder Mode [1]

“God is in the details. But I am not a creative person. I can’t draw, I can’t sketch, I can’t make anything. I just have to make sure things are being done right.” — Anna Wintour

“It is the responsibility of leadership to work intelligently with what is given and not waste time fantasizing about a world of flawless people and perfect choices.” — Marcus Aurelias

Lean in or sit back?

Your startup grew from 20 to 25 people this month. You don’t know if that qualifies you as a “Blitzscaler”[2], but you feel great about your new talent, especially your new Head of Sales, Agnes and your new Head of Product, Pablo.

The week they joined, you spent a few hours offloading your responsibilities to them. You walked Agnes through recent deals you won and lost, your sales playbook, and your best customer references. You trained Pablo on your product strategy and roadmap and your development process. You finished the week with a company happy hour to welcome them then spent the subway ride home gleefully deleting meetings from your calendar that you handed off to them. You fell asleep to dreams of hyperscaling.

It’s a month later, and in this morning’s leadership meeting, you are blown away by Agnes’ update. She flipped a set of dormant leads into active opportunities on the brink of closing. She’s made several upgrades to your sales playbook. She has a pipeline of new reps eager to join. You think, “This scale thing is easy!” and mentally pat yourself on the back.

Your love fest with your back stops at Pablo’s update. He shares some new projects he’s launching, and it’s feature salad. None of it connects with your startup’s strategy or goals or solves customer problems. You have an awkward discussion ahead of you where you’ll have to ask him to change the roadmap and pause the projects he just launched. You imagine his pushback: “Why did you hire me if you don’t trust me?”

You leave the meeting confused. You know that scaling your startup means hiring good people and delegating, but now you’ve been both rewarded and punished for doing just that. Don’t be too surprised, though, since all founders face this dilemma. It will help if you learn to Activate Founder Mode.

Micro-manage, macro-manage?

Conventional management wisdom is “hire great people and get out of the way.” That would seem to make sense for a growing startup, given how much work there is and how much talent you need to do it. Plenty of startups fail because their founders don’t learn to hire good people, train them, and trust them enough to take on important work.

Anyone who has worked at an early-stage startup knows this is incomplete advice, though. Founders struggle with which work to delegate and which to do themselves. They agonize about overruling decisions made by their team versus letting them stand. They debate which parts of their business they should go deep into and which areas to hand off. In my 30-year startup career, not a week has gone by that my teams didn’t struggle with these questions. It’s the core dilemma of managing at a startup.

Paul Graham’s Founder Mode essay, based on a talk by Brian Chesky, resonated with founders like me because it pinpointed that tension and gave it a name. We are told we should hire good people and delegate, but we see that fail constantly. We hate being micromanaged, but are tempted to micromanage others. We want to build our startup into a big company but know we can’t manage it like it already is one.

Unpacking these dilemmas requires remembering why a startup isn’t just a smaller version of a big company[3]. A startup is growing like crazy, changes every month, and is always adding new people. It has to constantly learn as competitors and incumbents respond. It has little room for error and can’t afford to waste much time or money. Most of all, it has to be as innovative at 100 employees as it was at 5.

Hiring people, handing them the work, and hoping for the best rarely solves these challenges, but micromanaging every person and project won’t either. As is usually the case, it’s not a binary choice.

Paul Graham doesn’t go deep on Founder Mode, but I can weigh in with my own definition. Founder Mode means learning where you should go deep and where you should sit back. It’s spending time with the right people and projects and letting others run on their own. It’s learning how to manage people and projects tightly but in a way that builds trust instead of eroding it. It’s accepting the risk of failure but not accepting sloppiness and low quality.

Founder Mode means learning to:

Pick your spots

Most of us want to be fair and treat people equally. If we have five direct reports, we assume we should manage them the same way, spend the same amount of time, and tell them their work is equally important.

We shouldn’t. People aren’t all the same, and the work isn’t all the same. You should spend more time and be more hands-on with the most critical projects and the least experienced people:[4]

If you are running out of money in three months, devote time to raising the round and work closely with the people who will help get that done, like your finance team. Don’t hesitate to tell other teams that their work is less important and that they’ll get less of your time, at least for now. You have to prioritize. Launching a new product is more important than optimizing your tech support. Designing your brand is more important than designing your office.

Your people are different and need different amounts of time and focus. Some will be new and be ramping up, where others will be more established. Some will have proven their excellence, and the jury will still be out on others who might not make it.

Find out where you are needed, put your time and energy there, and be willing to roll up your sleeves and go deep.

Build a Founder Mode team

Most startup management challenges have less to do with how you manage your team and more to do with who you hired in the first place, especially for executive hires, many of whom don’t work out and either quit or are asked to leave. Many of the startups I work with at Point Nine and elsewhere assume that up to half of their executive hires will fail, with many of them failing because they can’t work in Founder Mode.

A common failure path is to hire an executive from a larger company who can’t adjust to a startup. The executive builds a silo and runs their team with little input from you or from others. They aren’t hands-on enough to know if the work is any good, and their solution to problems is to ask for more budget to hire even more people.

Hire leaders who know their job is to get input from you and others and be willing to put their work up for scrutiny. Hire managers who are able and willing to activate Founder Mode themselves. They do exist; the best startups are full of them. You don’t want a team of managers who are too defensive to get feedback and too timid to give it.

Stay open to internal promotions, who often outperform outside hires even when they have far less experience on paper. They grew up in your company, are hands-on enough to contribute to the work, and are used to the discussion and debate that characterize a startup.

Front-load new hire training

Many founders, like the ones from our vignette, assume they should hire a senior person, spend a few hours training them, and then turn them loose, assuming anyone worth hiring can take over the job from there. This sometimes works, but it’s better to assume you have to make a big upfront investment in them.

If you hand off too quickly, you’ll get Pablo from our vignette, where a few weeks after the new executive starts, you get the sense that something isn’t quite right. Little work is getting done, and what does get done is low quality. Their proposed new hires make you scratch your head. They don’t know the details of the work. You now have to spend time and energy unwinding some decisions, and the team has lost faith in your new hire and in you for hiring them.

A better plan is to front-load the time you spend with a new hire and devote several hours a week for several weeks. Walk them through how you do things at your company. Shadow them on their first few projects. Sit in some meetings with them. Brainstorm solutions to problems.

They will get up to speed more quickly, and you’ll learn how to work together. You’ll also both discover if they are a fit for your company earlier. If they aren’t going to work out, both of you are far better off pulling the ripcord early.

Let the work guide you

Your startup has to deliver excellent work and deliver it quickly. As a founder, you are responsible for every piece of work that comes out of your company, whether you created it or someone you hired did. If you hire someone, delegate the work to them, and the work isn’t great, then you hired the wrong person, delegated too early, or didn’t delegate the right way.

Use the work to lead you to where you should spend your time. If the work from one part of your organization is excellent, congratulations. You probably don’t need to spend much time with that team other than checking in regularly and recognizing and rewarding them.

In the areas where the work isn’t excellent, you need to dig in. The people on that team might just need some more coaching and feedback. You might need to step in and help resolve some important decisions. In the worst case, you might have the wrong person running the team, and you have to ask them to leave and bring in someone new. But you can’t just sit back and hope things improve on their own.

Know your own strengths and weaknesses so you know which of your opinions you can be more confident in. But don’t limit yourself too much. Maybe you’ve never run a sales team before and your Head of Sales has, but you do know if you are making your plan or not, and you are perfectly capable of calling a prospect and asking why they didn’t buy or calling a sales rep and asking what help they need. You can also lean on your board and advisors in areas where you are weaker.

Talk to everyone

BigCo managers can mostly interact with their direct reports and a few peers. Some of this comes from big company politics, where managers want to control the information coming in and out of their team. “Stay off my turf, and I’ll stay off of yours” doesn’t work at a startup, though.

You need to know what’s happening in your own company. Find out how people are feeling and what questions and concerns they have. Test if they understand your strategy and plans. Find out who is performing well and who is struggling.

Have skip level one-on-ones. Talk to people working on important projects and critical deals. Have ad hoc conversations with whomever you bump into in the elevator or the lunch room. Encourage everyone who works for you to do the same thing.

If a manager at your startup is possessive and wants to “protect” their people from “being distracted” by anyone but them, that’s not a leader you want at your startup.

Be curious, not judgmental [5]

“Good people hate to be micromanaged” is accepted wisdom in modern management, but it has some nuance. When people complain about micromanaging, it’s usually about how that management is delivered. People will always remember how you made them feel, and all that. [6]

People bristle when feedback seems arbitrary or is delivered disrespectfully. If one of your product managers has interviewed 25 customers and put a hundred hours into designing a new feature, they won’t react well if you fire off a curt midnight Slack message telling them their solution is wrong and you know the right one.

Lead with curiosity instead. Ask what assumptions were behind the work. Ask what constraints they were under. Ask what else they considered. You might learn you don’t have all of the background. Maybe they have insight you don’t. They might be operating off of the wrong information or were never trained properly, which is your fault but something you can fix. You might learn that you’re just wrong, which should happen a lot unless you really are better at every job than every person at your startup, in which case you have bigger problems.

You will seldom hear someone complain about micromanagement if you treat them with respect, lead with genuine curiosity, consider the possibility you are wrong, and take the time to collaborate on the best solution.

Bake feedback into your operating system

Manger feedback also gets a bad rap because it’s often delivered reactively and unpredictably. You might have scars from a “Dilbert” manager at a previous job who shows up at a meeting for a project that is halfway done and dumps feedback on the team that they are on the wrong track. You may have been told the night before a customer meeting that you need to redo your slides. Everyone gets frustrated, the work needs to be redone, and the team flinches when they see you coming.

It’s more productive to bake feedback into the day-to-day cadence of running the company. Use your one-on-ones with your team members to regularly give and get feedback on their work. Schedule regular review meetings for important projects. Establish checkpoints for projects where you and other people will give feedback, and schedule them early enough in the project that work doesn’t have to get thrown out and redone.

You want to be known as a manager who helps teams do excellent work, not one who derails projects. Put in the investment up front to make that happen and tell everyone else you expect them to do the same.

Solve problems once

Sometimes, you’ll see work delivered by your startup that doesn’t meet your high standards. It might be a slideshow that’s ugly and misformatted. It might be a sales rep using out-of-date competitive intelligence. There might be more bugs in your product than you can accept.

When this happens, you get disappointed or angry, especially in moments of weakness when you are already tired or frustrated (which is why you should avoid the midnight Slack). It’s tempting to assume that someone screwed up and that you need to track down the culprit.

But it’s more productive to ask a few questions, find the root cause of problems, and propose more systemic solutions. Instead of getting mad at a sales rep for creating ugly slides, have your design team create a beautiful template and train the whole team on how to use it. If your competitive intelligence is wrong, assign a team to solve it and make sure they know to prioritize it.

Your reaction to all of this might be, “hmm…Founder Mode just seems like good management. Big companies should run that way, too. This is true. Although Apple could not have built the iPhone without thousands of talented managers and engineers, it also wouldn’t have built the iPhone if Steve Jobs had told that team, “go build a phone, and if you need any help, I’ll be in my office.”

Some will argue that only a founder like Jobs can operate in Founder Mode. Others say that anyone with enough fortitude can. It doesn’t matter. The point is that you can. Your startup probably wouldn’t even exist if the legacy incumbents in your market could do Founder Mode. Founder Mode is the only sustainable advantage you have, so never give it up.

Good luck!

[1] Paul Graham’s Founder Mode essay made the New York Times and has launched dozens of responses from founders and investors. You are reading one of them.

[2] The excellent Reid Hoffman book Blitzscaling covers what happens if your startup goes into hypergrowth.

[3] Steve Blank might have been the first to say “A startup is not a smaller version of a large company.”

[4] This is partially inspired by Andy Grove’s “Task Relevant Maturity” from his book High Output Management.

[5] Yes, this is from Ted Lasso, but you gotta admit that Ted’s a Founder Mode manager.

[6] Maya Angelou said, “I’ve learned that people will forget what you said, people will forget what you did, but people will never forget how you made them feel.” I know Angelou wasn’t a management consultant, but this is excellent advice for delivering feedback.

In the last few years, we’ve met a number of startups building carbon accounting, emissions management, and similar software products. Although all of these companies are working on an extremely important problem, I’ve been somewhat skeptical about many of the solutions I saw. There are two (connected) reasons for my skepticism.

1) Carbon accounting has a “Shit in, Shit out” problem

In most sectors, Scope 1 emissions (direct greenhouse gas emissions from sources owned by the company) and Scope 2 emissions (GHG emissions associated with the purchase of electricity, heat, or cooling) account for only about 25% of total emissions. About 75% of total emissions comes from Scope 3 emissions, i.e., the indirect emissions in a company’s value chain. In sectors like agricultural commodities, construction, and real estate, Scope 3 emissions represent around 90% of total emissions. And yet, according to a BCG estimate, only 10% of companies comprehensively measure their scope 1, 2, and 3 emissions. The rest rely on rough industry estimates for the bulk of their emissions. You can’t blame them — existing consulting services and software solutions just don’t enable companies to collect verified supplier-level data in a scalable, cost-effective manner.

So the vast majority of companies lack the data they need to properly measure and account for their GHG emissions. As the famous saying goes, “what gets measured gets managed”, and the inverse is true, too. If you can’t measure it you can’t manage it.

2) Carbon offsets can be (mis)treated as an “easy way out”

Many companies buy carbon removal or offset credits to improve their carbon footprint. While this is a good approach in theory (put a price tag on CO2 and let the market figure out who can reduce emissions or remove CO2 from the atmosphere most efficiently) it comes with major issues and limitations:

• First, many projects come with a huge amount of uncertainty. If you buy an offset certificate for one ton of CO2, how certain can you be that one ton of CO2 will indeed not be released into the atmosphere as a result of your credit? And if you buy a removal certificate, how certain is it that a ton of CO2 is permanently removed ? If you buy credits from a reforestation project, for example, you must be confident that the trees remain alive and healthy for several decades. That requires the forest to be protected against wildfires, droughts, pests, and other risk factors, which you can’t take for granted if we’re talking about a period of 30 years or more. Another issue is that it’s not always straightforward to verify the additionality of these projects, i.e., whether the emissions reductions or removals achieved by a project would have occurred without the financial support from carbon credit sales. This uncertainty is reflected in the price of carbon credits, which spans across a huge range (around $30-50 per ton of CO2 for reforestation, $400–1000 for Direct Air Capture).

• Second, there aren’t enough high-quality offset or carbon removal credits available to counterbalance the current global emissions. The idea that emissions reductions achieved in one area of the economy can compensate for emissions produced elsewhere makes sense, but getting to net zero requires the vast majority of companies to cut their emissions as close to zero as possible.

• Finally, by going the offsetting route, companies miss the opportunity to reduce GHG emissions by making improvements to their value chain (some of which might be cheap or even free).

As a result, I’ve always felt that there was a lot of greenwashing: You use an estimation of your emissions (which might be significantly off) and offset them by buying carbon credits (which might remove only a fraction of the CO2 they claim to do).

The need for reliable Scope 3 emissions data, and the huge difficulty in obtaining it, led Louis Coppey to develop strong conviction on industry-specific (AKA vertical) climate platforms long before I met the Root team. The thesis is that how you get Scope 3 emissions data — and what companies can do to reduce them — varies drastically across industries, making it impossible to build a horizontal software product that can be sold to various sectors without extensive service or customization. So when Eric and Maurice, co-founders of Root, told us that they’re building a climate platform tailored to agricultural supply chains, starting with a product that helps food and beverage companies collect primary data, it immediately clicked.

Enter Root Global

As you probably know, the food and agricultural industry is a major contributor to global GHG emissions. Around 25% to 30% of global emissions come from our food systems, rising to around one-third when we include all agricultural products. What’s probably less obvious is that more than 80% of the industry’s emissions occur with farmers and processors. Therefore, farm-level data is absolutely crucial to unlock reductions.

Collecting primary supplier data is time-consuming and expensive. A large FMCG brand might source milk, eggs, grain and other agricultural products from hundreds of thousands of farmers across the globe. In addition, they might source processed products like cheese or milk powder from thousands of food processors, and each of these food processors in turn might work with hundreds of farmers. This is where Root comes in.

Root’s software platform makes it easy for sustainability and procurement teams at food companies to collect verifiable primary data at scale. Farmers have to answer only a few questions, as most of the data that is needed for GHG emissions calculations is retrieved from existing documents and other data sources. Using this primary data, Root models each product’s environmental footprint according to the most up-to-date calculation methodologies, providing companies with the data they need to identify and act upon individual emission hotspots across their supply chains. Over time, Root aims to give precise indications of the carbon impact, cost, and timeline of reduction levers tailored to the entire agricultural value chain.

Root’s ambition goes way beyond data collection (as valuable as that is). By building a platform for all players in the food and agriculture value chain — farmers, processors, FMCGs, retailers — they’re creating a climate platform that helps companies across the food system to get to Net Zero. It’s an ambitious mission, but we can’t think of a better team than Eric, Maurice, Rodrigo, and their team of 24 Rooties to achieve it.

Our investment in Root Global

As I’ve mentioned above, my sense is that in the past, carbon accounting was often about estimating emissions and buying offers for marketing and PR purposes. I think we can and must do much better — by calculating emissions using primary data, setting reductions targets, and working towards them. And with increasing regulatory requirements, investor expectations, consumer demands (and temperatures), the time is now.

We’re super excited to announce our investment in the company (which has previously raised a pre-seed round from many good friends like Project A, P9 alumnus Robin Dechant and Cargo.one CTO Mike Rötgers) and are thrilled to work alongside (and root for) the team as they tackle this critical challenge. Root is currently looking for 10+ people across different roles. If you’re interested in joining them as one of their first 50 employees, please reach out!

A few days ago, Clio announced its US$ 900M Series F. The funding round was one of the largest capital raises for a vertical software company and the largest software financing round in Canada ever. Point Nine led Clio’s seed round in early 2009, so we’re fortunate to have been shareholders for most of the company’s history. When we invested, Clio had around 50 customers and a few thousand dollars in MRR. Today, Clio is used by more than 150,000 legal professionals, has crossed $200 million in ARR, is endorsed by more than 100 law societies and bar associations worldwide, and has more than 1,100 employees.

In this post I’d like to look back at Clio’s truly amazing 16-year journey and share a few learnings that might be relevant for founders or investors.

1) Some of the best companies start outside of the Bay area

It seems obvious now, but when Clio started out, the idea that world-class companies could emerge from outside the Bay Area wasn’t widely accepted. Back then, European and Canadian startups had a very hard time raising capital. A good example is Zendesk, where we didn’t manage to raise a Series A in Europe. Those days are, fortunately, long gone. The financing landscape has changed dramatically, making it much easier to raise capital from anywhere in the world.

Of course, there are still big differences between the SF Bay area and smaller tech ecosystems. The Bay area and other tech hotspots continue to produce amazing companies, and being in one of these regions still offers unique advantages (access to the largest talent pools, close proximity to early adopters, a rich ecosystem of experienced mentors,..). What makes many companies from other places so interesting is that because the founders are not part of the tech echochamber and because they are exposed to different types of industries and companies, they work on different problems.

Our portfolio company Laserhub, for example, has built the first fully integrated platform for the procurement of customized metal parts. Laserhub is based in Stuttgart, which is home to Daimler-Benz, Porsche, and thousands of small to mid-sized metal processing companies. It’s a multi-billion dollar industry, but if you haven’t grown up in an area like Stuttgart, it’s unlikely that you’ve had the type of industry expertise that led Adrian to start the company.

Similarly, no one in Silicon Valley was thinking about software for lawyers in 2008. The analogy with Laserhub isn’t perfect because there is no shortage of lawyers in California. But apparently, SaaS for lawyers wasn’t top of mind for Bay area founders at that time, maybe because vertical software wasn’t en vogue with Silicon Valley VCs in 2008.

2) Some of the best companies don’t follow the T2D3 path

For many years, investors were obsessed with the T2D3 growth path. When the 2022 downturn started, investors began to take a more balanced view, valuing efficiency over growth at all costs. However, hyper-growth is still the #1 thing that gets investors excited. Since returns are driven by power law, that’s understandable. Especially for larger funds, it’s hard to generate top-quartile returns without being in one of the few-in-a-generation companies like Wiz, which grew from 0 to $500M ARR in four years.

The reality is that very, very few companies can sustain this type of explosive growth over an extended period. Attempting to force it often leads to failure. As I wrote some years ago, the good news is that growing a little slower is not the end of the world. If you have a great product with high NPS, low churn, and an excellent position in your market segment, you have a decent chance of getting to $100M in ARR even if your growth rate starts dropping significantly below 100% year-over-year at around $10M in ARR. It just takes a few more years.

Except for the first few years, Clio never grew at 100% year-over-year. However, Clio has unusually high growth persistence, i.e. its growth rate didn’t go down much with increasing scale. In fact, the company even accelerated its growth in the last few years, disproving the Mendoza line idea, or at least being an exception to the rule.

While many hyper-growth companies have experienced massive slowdowns in recent years, Clio has not. One of the reasons is that hyper-growth was often fuelled by spending huge amounts on sales and marketing. So when these expenditures are cut back in a downturn, growth plummets. Clio never relied on excessive sales and marketing spend in the first place, so they didn’t have to cut back.

Another excellent example is Procore, the leading software platform for the construction industry. It took Procore 13 years to reach $10 million in revenue, so it was anything but a rocket ship. But then, in the following eight years, they grew from $10M to almost $900M.

3) The best companies all require huge persistence

I’m sure you’ve heard this many times before, but it’s worth repeating: the best companies all require huge persistence. Many founders would have given up before Procore reached $10 million in revenue, and surely most VCs would have abandoned a company with Procore’s early growth trajectory.

Not every startup idea is worth pursuing indefinitely. I’m not arguing for founders to keep going forever if things don’t work out. If an idea doesn’t work, you’ve pivoted once already, there’s no clear opportunity or market pull, it’s okay to give up. Life is too short. And don’t worry about losing money for your VCs; that’s baked into the model.

What I mean is that every company I know that made it big and looks like a huge success from the outside faced existential crises in the early days. And it doesn’t necessarily get better when you’re bigger. You just have different (and maybe bigger) problems.

A journalist recently asked me, “What made Clio win?”. My answer was that it began with Jack and Rian having the right insight at the right time. They knew what to build, for whom, and had the capability to build it. But ultimately, the #1 factor behind Clio’s success was their persistence. They faced numerous challenges and crises but never gave up.

4) Vertical software companies can become much larger than most people thought

For many years, most VCs didn’t want to touch vertical SaaS companies with a yardstick due to TAM concerns. What they’ve missed:

a) Higher market share

Vertical software companies usually face less competition, enabling them to capture a higher market share.

b) Ever-increasing ARPAs

Vertical SaaS companies can continually increase their average revenue per account through what’s become known as the layer cake strategy. What this means is that you continuously add not only new features, but new products, services and revenue streams. Louis goes into more detail here.

Clio began as practice management software for solo lawyers, offering a simple solution for time tracking, billing, and document management. Sixteen years in, it has become a true industry operating system, powering not every aspect of the legal process including client intake, client communication, court filings, accounting, and much more.

What’s more, Clio’s roadmap is as long as it’s ever been and there are no signs of deceleration at $200 million in ARR. Clio will keep adding layers to the cake, while also doubling down on its rapid market expansion upmarket and internationally.

Similarly, Procore, after more than 20 years, has captured less than 12% of its TAM in the US and less than 2% worldwide. The numbers are similar for other vertical SaaS winners like Toast, ServiceTitan, and Shopmonkey. These companies illustrate that the potential for vertical SaaS is much, much larger than most people thought 10–15 years ago.

5) The best companies write their own playbooks

Today there’s a well-understood playbook for vertical SaaS. You can find it on the Internet. ;-)

When Clio started, there was no playbook. Clio co-authored that playbook alongside a few other pioneering vertical SaaS companies of its generation. Similarly, when Zendesk began, there was no playbook for consumerized software and PLG. Zendesk, along with a few other SaaS companies that emerged between 2006–2009, helped create this playbook.

What does this mean for companies starting today? Founders should understand and leverage the playbooks that guided the success of the previous generation. Much of that knowledge remains highly relevant. But the best founders will go beyond these established frameworks and write the playbook for the next generation.

Don’t take this as advice to reinvent everything. There are many things I think founders shouldn’t try to reinvent. Don’t try to innovate when it comes to structuring an ESOP or sales compensation. Focus on innovation in product and GTM.

6) The best companies and investments are non-consensus and (eventually) right

You’ve probably heard it before, but I’ll mention it briefly because it’s so important. The idea goes back to Andy Rachleff, co-founder of Benchmark, who said that “in order to create something legendary, you have to have an insight that is non-consensus and right”. Imagine a 2x2 matrix. On one dimension, you can be right or wrong. On the other dimension, you can be consensus or non-consensus.

If you are wrong, you will fail, no matter what. But it turns out that just being right is not enough. There is one square you want to be in. And that is the square that is non-consensus and right.

Most people don’t realize that if you are in the right and consensus square, you will usually not achieve greatness. Your startup might have a good idea, but if it’s too obvious, multiple me-too competitors will get funded by me-too VCs. As competition floods the market, prices erode, and sales cycles lengthen. And the exit options become less attractive.

The path to greatness is to be non-consensus and right. Being non-consensus and right affords the startup the time to survive, adapt, and succeed after trial and error without fatal consequences. No one preys on them because no one believes their idea is important.

This gives the startup time to master differentiable and specific skills and build strengths for inevitable competitive battles that will come in the future. When you’re starting out, it’s way better if your potential competitors don’t care about what you’re doing.

Mike Maples, Jr. based on an interview with Andy Rachleff.

When we invested in Clio in early 2009, nobody wanted to invest. And vertical software was so unsexy that it took years before the first serious competitors emerged. Like Andy Rachleff said, this gave Clio the time to survive, adapt, succeed after trial and error, and eventually dominate the market.

7) The best tech companies are founder-led by a highly technical founder

Jack co-founded Clio with his co-founder Rian in 2008. Sixteen years later, Jack is still leading the company as CEO and he’s poised to continue for many more years, which is one of the reasons why we’re so excited about the future of the company.

There are exceptions, but based on our experience with companies like Clio, Zendesk, Jobber, Docplanner, and Brainly, we aim to invest in companies where (a) we think the founders can lead for many years or decades and (b) they are excellent technologists. One reason is that the best startups are so ahead of their time that even after ten or more years, they’ve only realized a fraction of their original product vision. To keep advancing, the founder’s vision is crucial. Another reason is that if there are sudden market or technology shifts (like with LLMs), you need leaders who deeply understand the technology and its implications and can adapt quickly. It’s hard for a hired CEO to do what Des Traynor did at Intercom when ChatGPT came out — refocus a large part of the company on AI, almost overnight.

8) Bonus learning: Always keep an eye on your spam folder. ;-)

Gladly co-founder and Point Nine Capital advisor Michael Wolfe is releasing a series of essays to help founders manage themselves and their startups. We’ll discuss founder motivation, productivity, mental health, relationships with co-founders, hiring and managing people, building a strong culture, and keeping people aligned and working on the right things. You can follow the full series at uninvent.substack.com.

In “Choose Abundance”, we’ll cover a pattern that emerges over and over when you start a company: “scarcity vs. abundance.” We’ll talk about how to recognize this pattern and how to make choices that are most likely to lead to success.

“If you want to go fast, go alone. If you want to go far, go together.” ― African Proverb

“Founders who consistently make decisions that build wealth are more likely to achieve what I call a “Rich” outcome (greater financial gains, lesser control), while founders who consistently make decisions that enable them to maintain control of the startup are more likely to achieve what I call a “King” outcome (greater control, lesser financial gains).” ― Noam Wasserman, The Founder’s Dilemmas

“If we don’t put aside our enmities and band together, we will die. And then it doesn’t matter whose skeleton sits on the Iron Throne.” ― Davos Seaworth, Game of Thrones

How much do you want it?

Your ex-colleague Marcy would make a perfect co-founder. She’s a technical wizard who has built products for your market. You are a sales and marketing operator with a pipeline of prospects interested in your product. You work well together and would make an ideal “hacker and hustler” co-founder dream team.

But you hesitate before sealing the deal. You’d give almost half of your equity to her, and you worry that she’ll slow you down. You know the strategy you want to pursue and don’t want to spend time debating it with a partner. You wonder if striking out as a solo founder would be more lucrative and less hassle.

Your bootstrapped startup is showing signs of life. Your revenue already covers you and your co-founders’ meager salaries. You can grow faster by spending more money, but you’d need to raise funding. A seed investor approaches you and offers you $2 million for 20% of your startup. She also insists you set aside 10% of the stock in a pool for future hires.

You keep the investor waiting for several days while your team debates what to do. Is the extra growth worth 20% dilution? Can you hire some people who won’t demand so much equity? If you increase your burn rate, will it lead to more and more financing rounds until you eventually own only a few crumbs of your own company?

You are choosing between two venture firms that have given you identical term sheets for your Series A. The partner leading the investment would join your board.

The first investor has a hands-off reputation. He sees his role as deferring to the founders and only intervening when asked. The second investor is more prominent and respected, but she is known for taking her board duties seriously and pushing founders to grow as fast as possible. A recent Hacker News thread trashed her for ousting a founder from his startup after she lost faith in his leadership.

You debate if it’s more important to have a heavyweight in your corner who will push you and your company harder, or would you rather maintain the freedom to chart your own course and not lose sleep worrying about keeping your job?

Your new COO has worked for you for six months, and she is fantastic — so fantastic that she would be a better CEO than you are.

After six years of hard work and stress, you are tempted to promote her and demote yourself to a smaller role running product management. You’d keep all your equity, work less hard, and have a higher chance of building a great company. But you wonder if you’d be satisfied not running the company that you built. What will it feel like to see her picture on the home page of TechCrunch instead of yours? How will you feel listening to podcasts featuring her talking about leading the company you started?

These kinds of tradeoffs pop up at every startup. They require you to ask how much trust you are willing to place in others and if you are willing to take leaps of faith. They require you to ask what you are willing to give up if you might get more back in return. They require you to choose abundance.

You gotta give to get

It’s easy to incorporate a company, declare yourself the CEO, own 100% of the stock, and make all the decisions. You only need to file a few forms and pay a small fee to the state of Delaware or your country’s equivalent.

But you own 100% of nothing. Turning that nothing into something requires a series of decisions that involve taking risks, giving things up, and putting faith in others. Adding investors and employees means diluting your stock. Building a strong board requires giving up some control. Adding senior people means sharing the credit and risking that they outshine you.

Most of these tradeoffs are some form of the “Rich or King” dilemma, best described in Dr. Noam Wasserman’s seminal book “The Founder’s Dilemma.”[1] You’ll also hear it described as “economics versus control” in the context of a round of funding.

I prefer to frame these in terms of the “scarcity vs. abundance mindset,” which describes how a person tends to approach important life decisions, especially those involving placing trust in other people.[2] (Thanks to my lovely wife, Mary, who alerted me to the concept.)

At a startup, scarcity vs. abundance manifests in most of the big decisions you’ll make:

Team — Do you bring in confident and experienced people who will ask for more stock and might challenge your decisions? Or do you recruit cheaper, more junior staff who will be more pliant and accept a role where they just do exactly what you tell them to?

Ownership — Do you accept that you’ll give up some of your ownership as you bring in investors, board members, advisors, and hire a team, or do you optimize for keeping the biggest piece of the company that you can, even if it means you get less help?

Idea — Do you stick with your original idea, even if customers aren’t excited about it? Or do you accept that your ideas aren’t as great as you thought and pivot until you find something that resonates, even if it’s not your original idea?

Delegation — Do you delegate important work to your team, even if they don’t do it as well as you would have (or worse, do it better)? Or do you approve all important decisions?

Co-founders — Do you add co-founders, even though you’ll give up a big piece of your ownership, share the credit, and risk co-founder conflict? Or do you stay a solo act and bet only on yourself?

Funding — Do you raise money to grow more quickly, even though that will dilute your stock and add more stakeholders with opinions on how to run your startup? Or do you grow more slowly but keep control of all aspects of the business?

Board — Do you build a strong board that will challenge you and your team and provide strong governance? Or would you rather go it alone and forego both the oversight and the help a board can provide?

Choose abundance, but wisely

My bias is probably obvious: I lean towards the abundance side of the spectrum. I know you have to take leaps of faith to build a successful company, even if that means having a smaller role and less control. I want to have the biggest possible impact on the market, build wealth, and bring a fantastic team along for the ride. It’s also more fun.

I’ve met founders who put their energy into “not getting screwed.” They obsess about dilution, giving the smallest equity grants possible when they hire people. They don’t trust investors, even ones with stellar reputations. They hire junior people who won’t challenge them, or they even try to outsource critical functions. I’ve never seen one of these founders do well.

But abundance doesn’t mean abandoning caution and being indiscriminate about who to trust and what risks to take. Abundance decisions, which usually involve who you want to invite into your startup, need to be made carefully. You need to do plenty of due diligence and take risks only when you have a good chance of getting more in return.

A few things you can think about are:

Know thyself and know thy startup

Some startups are capital-intensive and need funding and experienced staff to get traction. Founders of those companies should be realistic that raising money and adding skilled people to the company requires giving up some equity and control and brings along all of the hard work, stress, and messiness of managing a growing team.

Other startups have products that are cheaper to build and distribute. They can push off hiring and fundraising until further down the road. But even those founders are often too optimistic about how much money and talent it will take to remain competitive. A startup that needs less funding to launch also has fewer barriers to entry. Markets that grow quickly attract competition. Winning a market usually requires adding money and talent.

One way or another, you’ll face scarcity vs. abundance decisions, probably earlier than you think. Discuss them ahead of time with your team to ensure you are in sync about what kind of company you want to build and what you’re willing to give up to get it.

Face your “scarcity” demons

Human nature comes with the “loss aversion” bias, which is when the fear of losing something we have motivates us more than the opportunity to gain something we don’t have.[3] This bias was helpful a thousand generations ago when our ancestors were fleeing lions on the savanna, but it can slow a startup down.

Your life experiences also impact your scarcity vs. abundance mindset. Some founders tend to be more cynical about other people’s motives, less trusting, and thus less able to take the leaps of faith that building a startup requires. Other founders land too far in the other direction, where they are too trusting of others and don’t do enough due diligence when choosing who to work with.

We can’t psychoanalyze you, but you probably already know your natural tendencies. If you tend to approach situations from a place of distrust and spend more time worrying about what might go wrong than right, work on taking a step back and asking what you can do to get comfortable with abundance decisions.

Learn the intricacies of control

Some founders are reluctant to raise money or build a strong board of directors out of fear that their company will get “stolen” from them. This occasionally happens, but it doesn’t mean your only two choices are to never raise money or hand over the keys to your startup. You have plenty of options that lie between those two extremes.

Control is exerted along several dimensions, including operational control, financial control, voting control, and board control, each of which breaks down further into individual terms that are often part of the negotiation for a round of funding. Most of the terms you negotiate involve which decisions your investors and board will need to approve.[4]

We won’t get into the technicalities here, which are complex and vary by the type of company you are and where it is incorporated. You’ll need to work with your attorney in the context of a specific funding offer. But get familiar with the topic so that when the time comes, you can approach the tradeoffs with more nuance and have already thought about what you’re willing to give up and what you aren’t.

Keep control by deserving it

Many founders’ nightmare scenario is getting kicked out of their own company by their investors or even by their co-founders. It doesn’t happen often, but when it does, the story is almost always the same: the team has lost confidence in the founder’s leadership.

The loss of confidence usually starts when the startup starts missing its goals, but it usually takes more than that. Most startups go through difficult stretches, and the founders survive them as long as they are still seen as the best people to lead the startup. When it goes bad, it’s usually because the founders isolate themselves. They stop sending investor updates. They stop asking their investors and board for help. They don’t share the bad news with their teammates or enlist their help to solve it.

Even if you don’t have a board and investors who can kick you out of your company, you don’t have the control you think you do. You can’t force great people to join your company or to stay with you during hard times. If you don’t communicate and don’t trust your team, they’ll leave, and you’ll realize that you “control” less than you thought you did.

Do due diligence on team additions

You probably know how expensive a bad hire is, and we’ll certainly remind you in future chapters. A bad hire can slow you down, damage your culture, and heap a pile of stress on you. But you always have a simple remedy for a bad hire: you can ask them to leave.

You can’t ask investors to leave. They are with you for the long haul. Amazingly, founders often do more due diligence when they hire a new team member than when they add a new investor.

As you would for a key hire, spend as much time as you need to with a potential investor to get a feel for how well you’ll work together. Check lots of references from other founders who have worked with them, both ones the investor gave you as well as ones you’ve hunted down on your own. A good investor welcomes this and will respect you for it.

Choosing abundance helps set you up for a successful startup, but it’s no guarantee. You can increase your odds further if you Know Why You Get Paid (coming soon).

If you have feedback or suggestions for future posts, please comment or contact us at uninvent@substack.com.

[1] Noah Wasserman’s The Founder’s Dilemmas is a must-read for startup founders.

[2] Scarcity vs. Abundance was introduced in Steven Covey’s The 7 Habits of Highly Effective People. You may have also heard of Carol Dweck’s Growth Mindset, which is a related concept.

[3] Loss Aversion is another on our growing list of conceptual biases that trip up startup founders.

[4] Many resources are available on the specific items you negotiate in a term sheet, but Venture Deals is an excellent place to start. A good attorney will also help you understand the implications of the decisions you’ll have to make.