Airbnb Automates 60% of Code Under CTO Ahmad Al-Dahle

Since taking over as chief technology officer in January after leading Meta's Llama model development, Ahmad Al-Dahle has aggressively integrated artificial intelligence across Airbnb's $93 billion business. The shift has dramatically altered the company's engineering pipeline, with 60 percent of its codebase now authored by AI. This automation has helped the hospitality giant ship nearly 80 percent more features year over year, while boosting the average engineer's pull-request throughput by approximately 1.6 times.

To achieve these speeds, Airbnb abandoned traditional handoffs between product, design, and engineering teams, moving straight to interactive prototyping. The company relies on an internal context graph called Everest, which uses large language models, embeddings, and AI-based retrieval to map its codebase. Everest allows generalist developers to work across highly specialized areas. For example, while a grocery delivery service took eight to nine months to build, a subsequent airport pickup service took just six weeks because developers could leverage Everest to reuse previous organizational knowledge.

For practitioners, Al-Dahle's strategy highlights a pragmatic, multi-model approach. Airbnb deploys at least 10 customized models in production, selecting them based on a Pareto frontier of cost, performance, and latency. The company uses top-tier frontier models for coding tasks where errors are costly, but deploys smaller, post-trained models for latency-sensitive applications like search. This infrastructure also powers customer support, where AI agents now resolve nearly 45 percent of tickets, and an internal agent called AirChat that leverages Model Context Protocol.

Looking forward, Airbnb is deploying asynchronous agents running in containers to automate on-call engineering triages when monitoring systems like Grafana trigger alerts. Despite the heavy automation, Al-Dahle emphasizes that engineers must still understand their output. To preserve engineering craft, the company requires all developers to thoroughly explain any pull request generated by an AI before it is merged.