Ahmad Al-Dahle, who previously led Meta's Llama models, joined Airbnb as CTO in January and is pushing an "inside-out" AI strategy: first using AI to accelerate internal software development, then applying those capabilities to guest-facing products. The company reports that 60% of its code is now AI-authored, feature output is up nearly 80% year over year, and average pull-request throughput has risen 1.6x.
A key process change is moving product, design, and engineering teams directly to prototypes, with code itself as the artifact instead of separate requirement documents. Al-Dahle says this removes handoffs and is a major source of time savings. Customer support was the first user-facing deployment; AI now resolves roughly half of tickets, though the company deliberately keeps safety-related issues with humans. Synthetic data is used to test agents before production.
Airbnb's internal "Everest" context graph, built with LLMs, embeddings, and retrieval, lets generalists work across specialized code and sped up new services. Groceries took about eight or nine months to develop, while airport pickups, reusing learnings from the grocery project, took about six weeks. Airbnb describes itself as a multi-model company, using frontier and open models and doing much of its own post-training on open models.