Three operators. A few Anthropic licences. The throughput of a twenty-person team.
This isn't a forecast. It's what I've been observing in conversations with founders in Paris over the past few weeks, and again in Miami a fortnight ago. Engineering, finance, ops, marketing — entire functions are being run by a handful of people working alongside AI agents. The AI-native startups I'm meeting move at a velocity I've simply never seen before.
For two decades, when I met an entrepreneur for the first time, I asked the same opening question to size up the company: how many people are you?
It was a useful proxy. Headcount told me, roughly, what their ARR looked like, how fast they were growing, whether they were likely to be cash-tight, whether their KPI discipline was good enough to support the team they had built. It worked because, for thirty years, headcount and revenue scaled together in a reasonably predictable way.
I've almost stopped asking the question. It's no longer a useful proxy.
What I'm actually seeing
Three patterns repeat from one founder conversation to the next:
The same revenue, a fraction of the team. Companies hitting €1M, €3M, €5M ARR with teams of three, five, eight — figures that would have required twenty to forty people two years ago. The product is real. The customers are paying. The team fits in a single Slack channel.
Senior-only org charts. The work that used to be done by junior employees — research, data cleanup, drafting, testing, first-pass ops — is now handled by agents. Founders aren't hiring entry-level. They're hiring senior generalists who are good at directing AI systems and editing what comes out.
Speed as the primary moat. When a competitor takes six months to ship, you can ship in two. Iteration cycles compress. The feedback loop between customer signal and product change becomes daily, not quarterly. That speed is the new defensibility.
The proxy is dying — and the damage is broader than it looks
For investors, the immediate problem is diagnostic. If headcount no longer maps to revenue, what does? ARR per FTE — already a useful metric — becomes the dominant one. Customer concentration matters more, because three operators serving three large clients is a different risk profile than thirty people serving three hundred SMBs. The shape of the cap table matters more, because option pools sized for thirty-person teams are now overhang.
For incumbents, the problem is structural. A startup that ships at one-fifth your speed and one-tenth your cost base is not just a competitor — it's a redefinition of what's possible in your market. The big consultancies, the big agencies, the big software vendors all have to ask whether their cost-to-deliver is even competitive anymore.
For employment, the problem is societal — and not for me to solve in a single article. But it's worth saying out loud: when an agent can do the work of five people, the cost structure of every business changes. The barrier to entry collapses. The competitive landscape rearranges. And the question of how value gets distributed across the workforce becomes urgent in a way it hasn't been for forty years.
What founders should actually do about it
Three practical takeaways from the founders I think are getting this right.
Hire later. Hire senior. Hire generalists.
The instinct to scale headcount the moment funding lands is a habit from the previous era. It used to be a signal of momentum. Now it's a signal of sloppiness. The best founders I'm meeting today are deliberately under-hiring relative to their revenue — and using the slack to invest in tooling, agent workflows, and senior generalists who can each cover what used to be three roles.
Make AI tooling a budget line, not an afterthought.
The companies pulling ahead are spending real money on Anthropic, OpenAI, internal infrastructure, agent orchestration tools. We're talking five-figure monthly bills, sometimes six. That number looks shocking until you compare it to the four engineers you didn't have to hire. The math, in almost every case I've seen, is overwhelmingly in favour of the AI spend.
Re-architect the org chart, don't just bolt agents onto an old one.
The mistake I see most often: a company adds an "AI copilot" to an existing role and considers itself AI-native. It isn't. AI-native means the org was designed, from day one, around the assumption that agents handle a meaningful share of the work. That changes who you hire, how you structure teams, what processes you build, what your dashboard looks like. It's a different company — not a better-equipped version of the old one.
What I'm telling the founders we back
You can be AI-enabled, or you can be AI-native. The first is a feature. The second is a company.
Two things are happening at once. First, an enormous productivity step-up — the kind we last saw with the early web, and arguably bigger. Second, a quiet redefinition of what a "real" company looks like at every revenue threshold.
I've stopped asking founders how many people they are. I now ask: what does your team do that an agent can't? The answer to that question — clear, specific, defensible — is the new test of whether a company has thought hard enough about the world it's actually operating in.
You're seeing the same thing? Push back, refine, disagree. The shift is too fast and too consequential for any one of us to read it accurately alone.