AINEWS
Airbnb CTO Ahmad Al Dahle says 60% of the company's code is now AI written and feature shipping is up 80%, but the real shift is a workflow collapse, not a productivity trick.
Airbnb has killed the sequential handoff that used to move a product idea from spec to Figma to engineer to production test. The middle work, the part that took a week of tickets, mocks, and QA review, now belongs to AI tools. The company says the result is 60% AI-authored code, nearly 80% more features shipped year over year, and 1.6x pull-request throughput per engineer. (Ahmad Al-Dahle on Latent Space)
Those are Airbnb's numbers, not an independent benchmark. The roughly $93 billion public company is reorganizing its software org around that workflow collapse, and the CTO running it. (Latent Space)
The new pattern, which Al-Dahle calls "inside-out AI," starts with AI tools used internally to ship product faster, then pushes the same capabilities into the guest experience. In his telling, "inside-out" is the opposite of building a chatbot and bolting it onto a homepage. (Latent Space)
Under the old process, a product manager would write a requirements document, hand it to a designer for a Figma mock, hand that to an engineer for implementation, and then move the result to a separate QA step before launch. Each handoff added latency and a chance for the idea to drift. Under the new pattern, product, design, and engineering teams work from prototypes directly. The prototype is the spec. (Latent Space)
AI replaces the middle work: turning a design into a stub of working code, building the test harness, and writing the boilerplate. The engineer is no longer a typist translating a Figma file into a feature; the engineer is closer to a reviewer deciding which AI-generated draft ships. Al-Dahle frames this as a way to "deploy frontier models at scale into a core user experience, not just train them." (Latent Space)
The 60% figure doesn't mean Airbnb has replaced its engineers. It means the workflow around them has changed. The 80% year-over-year increase in shipped features and the 1.6x jump in pull-request throughput are second-order effects of that workflow collapse, not the cause. If the handoff goes away, more code can move through the same review surface. (Latent Space)
All three numbers are first-party. They come from a single interview with Airbnb's CTO, published on the Latent Space podcast, and they have not been independently benchmarked. There is no third-party engineer, no outside analyst, and no productivity audit in the source. A reader should treat "60% AI-authored" the way they would treat a CEO's claim that a new product line is the company's fastest-growing: a real signal, but a self-reported one. (Latent Space)
"AI-native company" is Al-Dahle's own framing, not an industry term with a settled definition. It is shorthand for: every team is expected to use AI tools to ship faster, and the org is being shaped around that expectation. Several other large public companies have used the same label in the past year. Airbnb is the clearest case so far of a company attaching three specific throughput numbers to the operational pattern in a single venue. (Latent Space)
Sequential product-to-design-to-engineering handoffs exist for a reason: they let each discipline push back at the spec stage. If a designer can show in Figma that a layout will not work on mobile, that feedback can land before an engineer has written a line of code. Compressing those steps into a single prototype means the friction has to move somewhere. Whether that review surface actually catches the problems a Figma review would have caught is an open question. (Latent Space)
Airbnb is a roughly $93 billion public company with a deep engineering bench and a product surface that is, at heart, a search-and-book flow. The inside-out thesis assumes the workflow collapse is a productivity gain, not a quality compromise. The first independent test will not come from a benchmark. It will come from a guest, or a host, noticing that a feature shipped faster than the old process would have allowed, and wondering whether the quality held. (Latent Space)
Al-Dahle has been Airbnb's CTO since January 2025. The Latent Space interview is the first public venue where he lays out the workflow change, the throughput numbers, and the "inside-out" framing in one place. What it does not yet contain is independent verification. That is the gap a reader should hold open. (Latent Space)