Why I'm leaving fintech

Late last year I was looking at an eighteen-month roadmap. Another pass at the underwriting model, the fourth one in two years, on a product line that had not materially changed since the second pass. Reasonable work, the kind any senior fintech engineer in Europe is doing right now. I had no opinion about any of it, which was new.

I have been working in European fintech since 2012, starting at Cringle and then through a few more companies in different roles. The work changed because the category kept producing problems nobody had answers to yet, and that kept me interested for more than a decade.

Fintech does not produce those problems anymore. There is plenty of work, most of it regulatory, but it is not the same work it was. The job is keeping the existing things running and surviving a regulatory environment built for institutions with balance sheets. It is important work and I do not want to do it.

This is not a downturn you wait out, because what is happening to fintech is not going to reverse.

The incumbents caught up first. The whole opening was an arbitrage on the fact that banks were technically late, and their mobile apps were so bad people would screenshot them on Twitter. Anyone who could ship adequate software had a position. Every major bank now has a mobile app that works, the neobanks won the new customers without taking the deposit base, the payments players ended up competing on price, and nobody is technically late anymore.

The second is regulatory load, and people want to point at one specific regulation as the villain, but the problem is older than that and a lot messier. The Single Market is a fantasy: every country reads EU regulations differently and adds its own rules on top, and the agencies inside each country read the same rules differently again. In Germany the agencies are risk-averse and will not decide on edge cases, so you wait months for answers a product team could reach in an afternoon, you build holding companies in three countries to passport licenses around so you can offer in country X what someone offers natively in country Y, and you produce paperwork nobody reads. Nothing ever gets removed, new layers just pile on.

The cost of all this is roughly fixed per company. Banks pay it out of free cash flow, fintechs out of product budget. I have watched compliance teams grow while product teams shrank or got redirected to compliance-adjacent features. Regulation favors incumbents not because the regulator is biased, but because the cost structure does not move with revenue.

You can start a fintech without much money and ship something, but you will hit the wall fast: licensing, passporting, compliance reviews, the agency that will not decide. It costs time and money, and experienced founders plan for both while the rest find out.

A SaaS company has none of this. Little or no regulation, fast iteration, cheap to build and cheap to run. If the upside is similar, SaaS wins on cost and speed every time. So builders with options pick SaaS, or AI, or anywhere with a lower friction floor, and fintech keeps the people who already have capital, balance sheets, or no other options. The innovation work left on its own.

Look at the outcomes. Revolut keeps marking up on secondaries, Klarna timed its IPO well, and the rest mostly look like firesales or limbo. Solaris took huge funding and got sold for a fraction, N26 has been stuck on its valuation for years with an IPO always somewhere in the future, and most companies in the middle either get absorbed for less than they raised or sit waiting for someone to pick them up. The work that comes out of all this is stitching the remains together: integration into the acquirer's core, compliance retrofits, the tenth iteration of a loan management system, modernization contracts for the incumbents who can finally afford to pay for them.

There is plenty of new fintech with AI right now, AI lending, AI accounting, AI for everything. Capital is flowing, but the underlying products are not changing, just getting cheaper and faster to run. The categories are the ones from a decade ago, fintech already ran this play with mobile and cloud, and LLMs are just the new layer. Startups will move faster than incumbents until they hit the same regulatory floor, and anything in regulated space now has to meet the EU AI Act on top of everything else.

The fintech company builders are the same dynamic made institutional. They exist to hand founders the whole regulatory substrate, because the cost of getting started does not work for individuals anymore and the substrate has to be amortized across many products.

The work fintech produces now is iteration, building better versions of known things in established markets. Iteration spans every function and the bar can be very high, but the test is different from innovation, where you are judged on whether the thing itself is new and useful rather than on how well you executed a known pattern. Most companies and most jobs are iteration, and most of the people doing that work are happy with it. If you want to build new things you have to be somewhere new things are being built.

Fintech still matters, it moves trillions of euros and people's money depends on it working, and the people running the late-stage winners built impressive things in the window when that was possible. All I am saying is that the window closed, what gets produced now is iteration, and if you want to build new things this is the wrong place to be.

Innovation work is strange because you do not fully know what you are building until you have built it. The categories are not settled, the conventions are still being argued about, and the tools you would want do not exist yet because nobody has needed them before. That is uncomfortable if you want predictable output, but it is the most interesting kind of work there is if you want to build things that did not exist before.

AI infrastructure is in that state right now. The basic questions about how agents find tools, what it means for models to compose, whether training works without a hyperscaler under it are still being figured out.

So I am out of fintech, into AI, and interested in the work again.