I'm writing this from San Francisco. I've been here about two weeks, and I fly back to London in a few days.

If you've been reading this newsletter, you'll know I came out here ahead of our seed round next year, to feel out what we'll be walking into. Two weeks in, I've had enough conversations, enough coffees, and enough late-night walks past packed founder cafés to know that something has genuinely shifted here. And whatever it is, it hasn't reached London yet.

The single most useful thing I've heard since arriving didn't come from a founder or an investor. It came from Garry Tan, the president and CEO of Y Combinator, who opened the AI Engineer conference last week with a talk I've been chewing on for days.

Now, I know YC is a divisive topic in the European founder scene. Plenty of people have strong opinions about the accelerator model, the terms, the culture. I'm not writing this to defend or criticise any of that. But there's one thing you can't really argue with. YC receives applications from tens of thousands of companies every single year. They see more early-stage deals than almost anyone else on the planet. And whatever they're seeing this quarter is what the rest of us will be arguing about in twelve months.

Whether you like their model or not, they are a barometer. And the reading they're giving off right now should make every European founder deeply uncomfortable.

See below to continue…

Join the waitlist for SoPhy - the networking intelligence engine!

Launching soon. Join the waitlist now.

Sign up here.

Continue…

Garry opened with a personal claim. In 2013, when he was a YC partner also writing code, he could produce roughly 14 usable lines a day. That was normal for a full-time engineer at the time. This year, he's measured his own output at about 400x that number. Same person. Fewer hours. Same models everyone else has access to.

The internet, predictably, dismissed the number as vibes. And honestly, arguing about whether it's 400x or 80x or 8x misses the point entirely. What's interesting isn't the multiplier. It's what happens to the shape of a company when even the floor is genuinely 8x.

Because if you're 8x faster than you used to be, you don't need the same headcount. You don't need the same coordination layer. You don't need middle management. You don't need entire departments whose only job was to coordinate the work of other people.

That's the shift I can't stop thinking about while I'm walking around here.

If Garry's personal stat is easy to dismiss, the batch-level numbers are harder to argue with. YC has publicly said that 25% of their W25 startups shipped products where 95% of the code was AI-generated. That batch went on to become the fastest-growing in YC history, with the whole cohort in aggregate growing 10% week over week, a rate that has never been recorded in early-stage venture before. Ninety-four YC-backed companies have now crossed $100M in revenue.

And critically, these companies are doing it with headcounts that don't compute against the traditional playbook. Emergent, an AI app builder from the summer 2024 batch, went from public launch to nine-figure ARR in eight months, sitting at 15 people when they hit $15M ARR. Retool is running at $60M ARR with roughly 40 people.

You can accuse YC of a lot of things. What you can't accuse them of is inventing the data. They're just reporting what they're seeing.

And what they're seeing is that a new company shape is winning.

Here's the part that's hit me hardest during this trip, and the one I want European founders to really sit with.

For twenty years, the default startup playbook has been the same. You raise a Series A. You hire 20 people. You hit Series B, you hire 100 more. You hit Series C, you're at 300. You scale a marketing team, a sales team, a customer success org, a product team, and an engineering org. And somewhere between the Series B and the Series D, you have between 200 and 500 people, and you're an actual company.

That whole middle band is now the worst place a startup can be.

Too big to move at the speed of a 15-person AI-native team. Too small to absorb the enterprise complexity that justifies a 5,000-person org. Too many people to run without middle managers. Not enough scale to make those middle managers economical.

And here's what nobody is saying out loud: the entire reason middle management existed was to coordinate human work. That was its only job. When work stops needing coordination in that shape, the layer stops needing to exist. And that's exactly what's happening.

The AI-native competitor with 15 people doesn't have a VP of Ops, three team leads, a project manager per squad, or a Slack channel called #sync-of-syncs. They have skill files, resolver tables, and eval rules doing the coordination work that a fifty-person layer used to do. And every dollar of revenue drops through to the bottom line at a rate the traditional org genuinely cannot match.

Which means the middle-band startup, the one at 200-to-500 people that used to be the healthy target for the industry, is now structurally the wrong company to be. Too expensive to run. Too slow to compete. Too big to pivot.

Think about what this looks like from the outside for a moment.

A mid-market SaaS company today. Let's say they're at $60M ARR with 400 people. Perfectly respectable. Growing 30% year over year. Preparing for their Series D. This has been the target company for the entire startup ecosystem for twenty years.

Now their AI-native competitor shows up. Same revenue. Forty people. Every dollar of revenue is dropping through at three times the margin. They can price 40% below the incumbent and still have better economics. They can ship product changes in days rather than quarters. They can enter three adjacent categories in the time the incumbent plans one.

What is that Series D going to look like when it lands in an investor's inbox next year?

This is not "AI will disrupt your industry" in the abstract futurist sense. This is a specific, already-happening problem for a specific band of companies, and the founders in that band are the last to notice because they're too busy running the business they already have.

Here's the part I keep coming back to as I sit in cafés out here.

In San Francisco, the founders who are building this way are surrounded by other founders building this way. They see it. They copy it. The investors they pitch to understand it. The comp benchmarks are being rewritten in real time. The peer pressure is doing half the work.

In London, in Berlin, in Paris, in Amsterdam, most founders are still building on the twenty-year-old playbook. Hiring aggressively into the middle. Raising to fund headcount. Building the middle layer that's about to become the least defensible position in software.

And they're doing it because the ecosystem around them is still measuring success by the old metrics. Headcount growth as a proxy for progress. Team size as a signal of ambition. "How many people are you?" as the second question every LP asks.

Meanwhile, YC's data is a preview of what the market is going to reward two years from now. And what it's going to reward is the exact opposite of what most European founders are optimising for.

The gap between the two worlds is getting wider every week, and I'm feeling it in real time being here.

If you're currently in the middle band, or planning to grow into it, the honest question is whether the company you're building is still the right shape. Not whether AI can make it a bit faster. Whether the org you're building is one that a competitor building from scratch today would ever choose to build.

For most European founders I know, if they're honest, the answer is no.

The rational move for a lot of founders right now isn't to scale. It's to consciously stay small, wire the work differently, build the company brain, and let your revenue per head compound in a way it never could before. If you're at Series A and thinking about hiring twenty people, the question is whether ten of them should be skill files instead.

If you're pre-seed like me and thinking about your team over the next twelve months, the answer is even clearer. You are the last generation of founders who gets to choose what shape their company will be. Every founder who came before you had to build a middle. You don't. You get to skip the layer that's about to become an active liability.

And I think the ones who see this shift first, who build the AI-native company from day one rather than the traditional company retrofitted with AI, are going to be genuinely difficult to catch. Not because they're smarter. Because they're not carrying the coordination cost of a shape that no longer works.

I came to San Francisco expecting to leave with a list of investor names and a rough plan for our seed. I'm leaving with something more useful and slightly more uncomfortable, which is a real question about the shape of the company I'm building.

I don't know if Garry's 400x number is exactly right. I don't need it to be. Because whether the multiplier is 400x or 80x or 8x, the direction of travel is the same. The companies that win the next decade are going to look nothing like the ones that won the last one. And YC, whatever else you think of them, is reporting from further inside that shift than anyone else in the industry.

The company shape most European founders are still building for is the one that's about to disappear.

The uncomfortable question is what we do about it when we get home.

✅ Know a founder currently scaling into the middle band and not thinking about what that might mean two years from now? Forward this their way. The sooner they ask whether the shape of their company still makes sense, the better the decisions they'll make between now and their next round.

(👉 Vote now — we’ll share the results in next week’s issue. All votes are anonymous.)

(Last week’s poll results)

All the UK’s best startup events in one place.

Access here and please share it with others.