Airbnb says AI writes 60% of its new code, and now every board has a number to ask about

Brian Chesky told investors on Airbnb's Aug. 6 earnings call that AI is now writing 60% of the company's new code, and that Airbnb is building, testing and iterating faster than it could a year ago ^[1][2]. The number matters less for what it says about Airbnb than for what it does to everyone else: it is portable, quotable, and a board member can now put it in front of any engineering leader and ask why their figure is lower.

The rest of the disclosure is the part worth auditing. Airbnb reported $3.6 billion in second-quarter revenue and beat expectations ^[3], and Chesky said he plans to spend more on AI tokens this year as part of a push to become an "AI-native company" ^[4]. He described monitoring employee progress on AI initiatives, token usage, and the output of the AI itself ^[5]. The customer service tool resolved 45% of user inquiries without human intervention ^[6], the company plans to test AI search ^[7], and Chesky said AI is accelerating work across search, sign-up, checkout and payments, converting more traffic into bookings in what he called one of the biggest drivers of growth ^[8].

Two cautions. Fortune's report attributes the 60% to Chesky without specifying what is being counted, lines or commits or pull requests, or how it is verified ^[9]. And earlier this year Chesky told investors AI had cut time from concept to launch by as much as 60% ^[10]. Two different 60s, one an input and one an outcome, are easy to blur into a single impressive story. The share of code an assistant produces is a measure of tooling adoption, not of throughput; the 45% resolution rate is the sturdier number because a customer either got an answer or did not ^[6].

The capital-efficiency read is where this stops being a management-philosophy story. Airbnb has about 8,200 employees ^[11] against a roughly $110 billion valuation ^[12], which works out to about $439,000 of revenue per employee in a single quarter ^[1] and about $13.4 million of market value per employee ^[2]. That is the actual claim behind founder mode: not that the CEO enjoys being in the weeds, but that a founder watching token spend and shipping cadence at that level of detail is running a cost discipline, not a personality.

Founder mode entered circulation through Paul Graham's 2024 essay, itself prompted by a Chesky talk at a Y Combinator event ^[13]. Chesky's own framing to CNBC was that the winners of AI will be companies that embrace transformation, "founder mode, not manager mode" ^[14]. He has since qualified it: start in the details, develop trust and muscle memory, then let go, because "great leadership is presence, not absence" ^[15]. The template does not scale easily. By Fortune's count, 26 of the 2026 Fortune 500 are founder-led ^[16].

What to watch is whether the input metric turns into revenue. Chesky is launching his own AI lab, with details still thin ^[17], and says the industry is at the start of what he calls a thousand-year journey ^[18]. He also expects up to 18 months before the new AI software debuts and has called the path a "very, very tall mountain" ^[19]. Meanwhile, people close to the matter told Bloomberg that turning Airbnb into a one-stop travel shop could add $1 billion in revenue ^[20], roughly 28% of a single quarter's current top line ^[3]. If that lands, the 60% was worth quoting. If it does not, it was a productivity statistic with no denominator.