On Airbnb's Q1 2026 earnings call on May 8, CEO Brian Chesky said 60% of the company's new code is now AI-generated. Managers are expected to write code and use Claude Code directly. The customer support AI bot went from handling about 33% of issues to 40% in Q1 2026. Three data points, one earnings call, zero methodology footnotes.
The specificity of an earnings call matters here. This is not a developer conference keynote where "AI writes most of our code" gets a standing ovation and nobody checks. It is a regulated disclosure where Chesky is accountable for what he says to investors. That makes the 60% figure the most credible public benchmark for AI coding adoption in production from a major tech company - credible, not verified. Nobody published the spreadsheet.
For scale: in 2025 Sundar Pichai put the AI-generated share of code at Google at over 30%. Airbnb's 60% is roughly double that baseline.
What Chesky actually said
"AI gives huge leverage - where you might have needed a team of 20 engineers before, an engineer can now spin up agents to do a lot of work under supervision."
The framing is force multiplier, not replacement - agents as supervised workers, not autonomous actors. So far, so standard executive AI script. The more interesting point from the call: Chesky admitted that conversational chat UI does not work well for Airbnb's core use case. Travel search, group bookings, direct object manipulation, comparing listings across a city - these do not reduce well to a chat thread. You do not book a weekend in Lisbon for four people by typing into a box.
His team uses AI to write 60% of new code, but they are not building a chatbot for customers. A CEO who can tell the difference between where AI helps and where it does not is rarer than the 60% number suggests.
AI support adoption trend
The customer service progression from 33% to 40% handled by AI is a more reliable signal than any percentage about code generation. Support deflection is measurable, has a clear unit (issues handled without escalation), and is tied to real cost savings. Code generation percentages depend heavily on what you measure - lines, files, commits, tokens, or editor autocomplete acceptances all produce different numbers.
The methodology problem
Airbnb has not disclosed how it measures "AI-generated code." This matters because:
- GitHub Copilot autocompletions accepted = one definition
- Full agent-generated files with no human editing = a much stricter definition
- Anything in between covers an enormous range
Amazon dropped Q Developer last week in favor of Claude Code (covered here). Same week, Airbnb reports 60% AI-generated code. The pattern is real. The precision of the number is not.
The manager-coding mandate
The instruction that managers must now write code or use AI coding agents is the most loaded sentence from this earnings call. It is an explicit renegotiation of what an engineering manager's job is - or a pressure valve dressed as a job description. Making managers "code with agents" sounds transformative until you picture a VP pasting a Jira ticket into Claude Code at 11pm and calling it engineering leadership. Whether this produces better product decisions or just adds one more skill tax to a role already stretched thin between IC expectations and people management is a live question. The earnings call did not answer it. Neither will the next one.
Q1 2026 financials were solid: $2.7B in revenue (+18% year over year), $160M net income (+3.9%), 156.2M nights booked (+9%). Airbnb has been cutting headcount since 2022.