The fragmentation dividend
AI has collapsed the cost of writing software. The one thing it can't shortcut is knowing the customer — and Europe has been forced to practise exactly that for decades.
Paper collage. A photograph of the Atomium in Brussels, cut into diagonal strips and reassembled over a grid of numbers.Image: Anna Riepe & FARI / Better Images of AI / CC BY 4.0
Last week at my company, the AI Aladdins on their vibe-coded magic carpets launched a word-processing tool that lets you collaborate in real time with your AI agent. After being asked multiple times if I was "absolutely loving it" for editing articles, the answer was still "No." Meanwhile, the engineers and ops people who were using it to spin up internal briefing and spec notes — modern paperasse, or busywork — were obsessed.
The tool isn't useful to me because no one asked, "As someone who uses Google Docs all day long, what kind of new word processor would be helpful to you?" No one shadowed the way I work or watched my screen. They built for the user they imagined, not the user they had.
That's a small story, but stock pickers might want to heed it. AI has collapsed the cost of writing software. Anyone with a frontier model and a weekend can now build the generic horizontal product that used to require a Series A. So while a lot of generic software is cooked, one flavour will thrive: software that knows something. The specific customer. The specific industry. The specific regulatory wrinkle that took someone five years of sitting in meetings to understand. That kind of knowledge is the moat — and it's the one thing AI can't shortcut, because nobody has written it down.
Building software like that turns out to be something Europe has been forced to practise for decades.
No choice but to specialise
European founders had no choice. You cannot build a single dominant "European" HR platform when your users operate under 27 different regulatory regimes in 24 official languages. You build for Germany, or for France, or for Dutch SMEs specifically — and you learn those customers in granular detail, because there's no shortcut around the rules.
Look at neobanking, the canonical example of fragmentation-as-strategy. Six of the world's ten most valuable neobanks are European — Revolut, Wise, Monzo, N26, Starling, Bunq — because the fragmented landscape demanded multiple specialised champions instead of one winner-take-all platform. N26 in Germany. Bunq in the Netherlands. Each of them had to know their home market in a way a generic challenger never would.
A former colleague who just raised $40m for a vertical-SaaS rollup made her first investment last month in an exciting, fast-growing UK software company that makes software for petting zoos, indoor skydiving facilities, and hay rides. Please don't tell me the hay-ride operator is going to vibe-code something to better capture group reservations. They wouldn't know where to start, and they certainly can't pay for enterprise-grade alternatives.
Wealth manager Jan Voss put the argument cleanly: software tailored to particular needs would also fuel "the small businesses behind these bespoke software solutions rather than massive corporates with bloated overhead."
So let's hear it for an ERP built specifically for mid-sized German manufacturers, or a compliance tool that understands French employment law. These tools require someone to have sat with the customer and watched them work — or to have been the customer themselves, feeling the pain. They cannot be vibe-coded in a weekend. Even Amazon's engineers, with all their resources, learned this the hard way this month: AI-assisted infrastructure changes contributed to a six-hour outage at one of the most technically sophisticated companies on earth. If hyperscalers can't fully trust AI to ship infrastructure, the regional manufacturer in Stuttgart certainly can't trust it to ship a payroll system that complies with Betriebsverfassungsgesetz.
The tradeoff — and why it matters less than it looks
There's a tradeoff to the "fragmentation as dividend" argument. You don't build a Google-scale company on hay rides. Investors will argue this means fewer tech millionaires going off to start the next wave of businesses. But the counter is sharper than it first looks. In an AI era, hyperscale advantages get commoditised fast — generic software is exactly what frontier models can rebuild cheapest. What survives is the stuff that's hard to replicate because it's embedded in customer relationships and domain knowledge nobody has written down.
Europe's old weakness — no winner-take-all platforms — is starting to look irrelevant. Europe's old strength — defensible vertical businesses — is starting to look like the only kind of software business worth owning.
In software, as in most things, the people who couldn't cut corners often end up with the better product. Europe may have been, inadvertently, training for this.
The bottom line: the skill about to become disproportionately valuable isn't engineering — it's the patience to sit with a user and watch them work. AI has collapsed the cost of writing code, which means the bottleneck has moved to knowing what to build. Product > model.
Continue reading
This column is for Second Chapter members. Join the executives, founders and policymakers reading our reporting from inside the room.
From €250/year. Become a member to read in full.
The conversation
14 comments · 3 from our teamSecond Chapter is a multi-directional resource. Comment, challenge our reporting, or suggest an angle for the next piece in this series — our journalists are reading.
The hay-ride line is doing a lot of work here, but it's the right instinct. If you're building vertical software for an "unsexy" market and AI has changed your moat — for better or worse — I'd love to hear how. Reply here or message me directly.
This matches our thesis exactly. We've quietly shifted toward vertical SaaS in regulated niches precisely because the domain knowledge is the defensibility. The "fewer tech millionaires" worry is overblown — a portfolio of defensible €50–100m exits beats one moonshot that gets commoditised.
Well put, Mathilde. The portfolio-of-moats point deserves its own column — flagging it to the Advisory Group for the next theme cycle.
"Product > model" should be on a poster in every European founder's office. The patience to sit and watch a user work is genuinely undervalued right now. We hired two ex-operators from the industry we sell into and it changed everything.