Airbnb Says AI Writes 60% of New Code, but Complex Products Still Need Real Interfaces

Airbnb CEO Brian Chesky said AI now writes about 60% of the company’s new code, according to a May 8 TechCrunch report. He also said Airbnb’s AI customer service agent is handling around 40% of support work. Those numbers are striking, but Chesky’s broader point may be more important: complex consumer products like travel and e-commerce will not simply collapse into chatbots.

The Airbnb example shows how AI is actually entering large technology companies. The loudest AI product narrative often imagines a universal chat box where users type a request and the system handles everything. But travel products are more complicated than that. Airbnb has listings, photos, maps, prices, dates, reviews, cancellation policies, host communication, payments, disputes, safety issues, and local context. Users do not just need answers. They need comparison, trust, visual inspection, and structured decision-making.

That is why AI may first have its biggest impact inside the company. Code generation is one clear area. If AI can produce boilerplate code, tests, migrations, documentation, internal tools, and simple feature work, engineering teams can change how they operate. Developers spend less time writing every line manually and more time defining requirements, reviewing output, designing architecture, validating behavior, and handling edge cases.

Customer support is another obvious use case. Travel platforms receive large volumes of repetitive questions about bookings, refunds, check-in rules, account access, cancellations, and policy details. If an AI support agent can handle 40% of support work, it is likely absorbing a large share of routine, information-heavy cases. Human agents still matter for serious disputes, safety incidents, payment problems, legal issues, and emotionally sensitive cases.

The product design lesson is important. AI does not necessarily replace interfaces. In many cases, it improves them. A traveler might ask in natural language for “a quiet place for a family of three near public transit under a specific budget,” but the final decision still benefits from a map, photos, filters, reviews, price breakdowns, and house rules. AI can narrow choices, summarize trade-offs, explain policies, and generate itineraries, while the interface provides structure and trust.

This is relevant far beyond Airbnb. Many consumer technology companies are trying to decide whether they should rebuild their products around chat. The more realistic answer may be hybrid design: AI for intent understanding, search, summarization, support, and personalization; structured UI for browsing, comparison, confirmation, and transactions.

The 60% coding figure also deserves careful interpretation. It comes from an executive statement, and outside observers do not yet know the exact measurement method. It could refer to generated lines of code, new code touched by AI tools, accepted suggestions, or code produced with human editing. AI-written code is not the same as AI-owned software quality. Testing, review, security, performance, architecture, and product judgment remain human responsibilities.

Still, the direction is clear. AI coding has moved beyond experimental side projects and into the production workflows of major internet companies. The same is true for AI customer service. The next question is how companies measure the quality, reliability, and long-term maintainability of AI-assisted work.

The takeaway: Airbnb’s numbers show that AI is already changing how software and support work get done. But its product comments are just as important. The future of consumer apps is unlikely to be “everything becomes a chatbot.” It is more likely to be AI layered into well-designed interfaces that help users make complex decisions.

For product teams, this is a useful corrective. Chat is powerful for expressing intent, but it is not always the best surface for evaluation and purchase. Users often need to compare options visually, scan details, adjust constraints, and feel confident before paying. In travel, that confidence comes from structured information: dates, maps, photos, reviews, fees, policies, and availability. AI can help organize those details, but it should not hide them.

For engineering teams, the Airbnb example also raises questions about how AI-generated code is managed over time. If a large share of new code is AI-assisted, companies need strong standards for review, testing, security, documentation, and maintainability. The risk is not that AI writes code. The risk is that teams accept code they do not fully understand or cannot maintain later.

The next things to watch are whether Airbnb discloses more detail about how it measures AI-written code, how its AI support agent affects customer satisfaction, and how AI features appear in the user-facing booking flow. The most successful consumer AI products may be the ones that make complicated decisions feel simpler without removing the structure users need to trust the result.

Source: TechCrunch