Airbnb's CEO Says AI Writes 60% of Its Code — and 'Founder Mode' Is How You Survive It

Airbnb CEO Brian Chesky said in a Fortune interview published Thursday that artificial intelligence now writes 60% of the code at the company — and he argued that the only management model capable of handling that pace of change is “founder mode,” the hands-on style he has championed since Paul Graham coined the term in 2024.

The 60% figure isn’t brand new — Chesky first disclosed it on Airbnb’s Q1 2026 earnings call in May. What’s new is the framing: two years into the AI coding revolution, Chesky is staking out a position that the technology doesn’t just change how software gets written. It changes who should run companies, how closely they should watch the work, and why the conventional playbook of delegation may be actively dangerous right now.

The numbers behind the claim

Sixty percent of code written by AI sounds like a stat about engineering productivity, and it partly is. On the earnings call, Chesky told investors that AI helped Airbnb ship new improvements faster and cut the time from concept to launch by as much as 60%. Engineers are using AI coding tools to build products faster and with fewer defects, and the company is testing natural-language, conversational search for bookings on the consumer side.

But the more revealing detail is who is using the tools. Chesky said managers at Airbnb are also “getting their hands dirty” with coding — many of them using Claude Code, Anthropic’s command-line coding agent. That’s a striking inversion of the traditional division of labor, where managers write reviews and engineers write code. At Airbnb, the expectation now runs the other way: if AI has collapsed the cost of writing code, then understanding code — directing it, reviewing it, sitting inside the loop — becomes part of every leader’s job, not just the ICs.

This mirrors what other large tech companies have reported. Google and Microsoft have both said AI generates a substantial share of their new code, and startups routinely cite higher figures. Airbnb’s disclosure matters because it’s a consumer internet company — not an AI lab — with a 3,000-plus workforce, showing that AI-first engineering has become standard practice at mainstream scale.

What ‘founder mode’ means in the AI era

The term “founder mode” comes from Paul Graham’s influential 2024 essay, which argued that the conventional advice given to founders — hire good people and get out of the way — often fails, and that great founders stay in the details. Chesky became the movement’s poster child for running Airbnb with unusual directness: small teams, direct engagement with work at every level, minimal layers between him and the product.

In the Fortune interview, Chesky applies the doctrine to AI specifically. He described founder mode as closely monitoring employees’ progress on Airbnb’s AI initiatives, their token usage, and the overall output of the AI tools themselves. That last part is the tell. When a majority of your code is machine-written, the unit of management isn’t the person anymore — it’s the system: which models are being used, how many tokens they consume, what the agents produced, and whether it’s any good.

Token usage as an executive metric is a genuinely new idea, and it may be the sharpest one in the interview. Compute is now a line item that behaves like a raw material — per-employee, per-team, per-feature. A CEO who can see it can steer it. A CEO who only sees quarterly rollups cannot.

Chesky’s warning, as he’s framed it before, is directed at managers who refuse to engage: leaders unwilling to get into the details of AI tools risk becoming irrelevant, because the people they manage are suddenly 60% more productive while their own judgment stays calibrated to a slower world.

The counterargument worth taking seriously

It’s easy to read founder mode as a defense of micromanagement with better branding, and critics have said exactly that since Graham’s essay appeared. The failure mode is obvious: a founder who inserts themselves into every decision becomes the bottleneck, and Airbnb’s own history includes periods where Chesky’s closeness to the product slowed things down.

But the AI-era version of the argument has a firmer foundation than the original. When the technology stack shifts this fast — new models, new agents, new pricing, new failure modes every quarter — institutional knowledge decays quickly, and the leaders closest to the tools have the freshest information. Delegation works when the environment is stable enough that last year’s playbook still applies. Right now, it doesn’t. Chesky’s bet is that founder mode is simply the rational response to an environment where the ground moves every 90 days.

There’s also a talent dimension. Engineers who want to work at the frontier are drawn to companies where leadership genuinely understands the tools. A CEO who can discuss token economics and agent output credibly is a recruiting asset; one who can’t is a signal that the company will be a follower.

Why it matters

The interview crystallizes a shift that’s been building all year. The first phase of the AI coding story was about individual productivity — developers shipping faster. The second phase, which 2026 has ushered in, is organizational: what happens to management layers, headcount planning, and the definition of an engineering team when most code is generated.

Airbnb is offering one answer: flatten, inspect closely, and let leaders code. Whether founder mode scales — or burns out the founders practicing it — remains the open question. But with 60% of code now machine-written at a $100-billion-plus public company, the debate is no longer theoretical.