Airbnb has put a hard number on AI-assisted software work, saying AI now writes about 60% of its new code. The bigger signal is that coding is no longer being treated as work only for engineers.
Airbnb CEO Brian Chesky has joined the growing list of tech leaders turning AI coding into a boardroom metric. During the company's latest earnings discussion, he said AI wrote about 60% of the code Airbnb engineers produced in the quarter, while managers are being pushed back toward hands-on building through tools such as Claude Code.
That puts Airbnb between Shopify, which has cited roughly 50% AI-generated code, and Google, where Sundar Pichai recently said 75% of new code is AI-generated and then reviewed by engineers. The numbers are not perfectly comparable, and that matters. But they are still useful because they show how quickly AI coding has moved from an experiment inside engineering teams to a management priority inside some of the world's most closely watched technology companies.
As TechCrunch reported, Chesky framed AI as leverage for teams building tools for Airbnb's API partners, the software providers that help hosts manage properties across different systems. His point was practical: work that once required a larger engineering group can now be pushed forward by a smaller team using agents under supervision. For founders, that is the part worth studying. This is not just a story about autocomplete getting better. It is about companies rethinking how much software they can ship, who gets to build it, and how many layers of coordination they really need.
The most interesting part of Airbnb's message was not the 60% figure by itself. It was Chesky's comment that the company has less room for pure people managers and that design and engineering managers are getting back into coding or using Claude Code directly. That is a sharp turn from the familiar model where senior managers move farther away from the product as they gain scope.
There is an obvious operating advantage here. A manager who can turn a product idea into a working prototype, review an AI-generated change, or understand why an agent is going down the wrong path can make faster decisions. The old handoff chain gets shorter. A product leader does not need to wait for three meetings to know whether an idea is technically awkward, trivial, or dangerous.
That can be especially powerful inside startups. Small companies already depend on people crossing job boundaries. AI coding tools make that habit more valuable, because a founder, designer, operations lead, or customer support manager can now create internal tools that would have sat in the backlog for months. The advantage is not that everyone becomes a full software engineer overnight. The advantage is that more people can participate in the first draft of software.
But the same shift creates governance risk. Managers who can generate code are not automatically equipped to judge architecture, security, privacy, or long-term maintainability. A quick internal tool can become a production dependency. A small workflow automation can start touching customer data. If AI makes it easier for non-engineers to ship, companies need clearer rules for what can be shipped, what must be reviewed, and who owns the result after the first version works.
The percentage claims need context
The phrase AI-generated code sounds precise, but it often hides more than it reveals. One company may be counting generated lines accepted into a codebase. Another may be counting AI-assisted commits. Another may include refactors, tests, boilerplate, migrations, or code produced by agents and then heavily rewritten by humans. Without a shared definition, 60% at Airbnb and 75% at Google should not be read like identical productivity measurements.
There is also a difference between new code and the full codebase. Airbnb's claim appears to be about code produced by engineers during the quarter, not a statement that 60% of all existing Airbnb software was created by AI. That distinction matters because older systems, core infrastructure, payment flows, trust and safety tools, and marketplace logic may still depend heavily on human judgment and institutional memory.
The better way to read these numbers is as a signal of workflow adoption. AI is now present in a large share of software creation at major tech firms. It drafts, rewrites, tests, migrates, and scaffolds. Humans still review, constrain, and take responsibility. The strategic question is whether companies can turn that workflow into better products rather than simply more code.
Airbnb has already tied AI to customer support, saying its assistant can resolve more than 40% of issues without a human agent. That is another useful clue. The company is not only using AI to make engineers faster. It is also testing whether AI can change service costs, partner tooling, and eventually the travel experience itself. Chesky has been cautious about claiming that anyone has solved AI for travel or e-commerce, which is the right caution. Booking a stay is not the same as summarizing a document. People compare, hesitate, ask messy questions, and care deeply when something goes wrong.
For founders, the takeaway is simple but uncomfortable. AI coding tools are becoming part of the operating system of modern companies, not a side project for engineering teams. The winners will not be the companies that boast the highest percentage of AI-generated code. They will be the ones that combine broader software creation with disciplined review, stronger technical ownership, and managers who understand enough of the work to make better calls. The next metric to watch is not how much code AI writes, but how much useful product reaches customers because of it.
Also read: AI diplomacy is becoming a market risk for startups • AI data center noise is becoming a neighborhood fight • Meta's email mishap shows why AI agents need real controls
