Airbnb CTO Ahmad Al-Dahle: AI resolves half of support tickets

Al-Dahle joined Airbnb as CTO in January after serving as head of generative AI at Meta, where he led the launch of its open source Llama models over 2023-2025. His stated reason for the move was to chase the next hard frontier: at Meta, he said, "we kind of knew what the flywheel looks like," while the open challenge now is deploying models at scale in ways that add value to a core user experience. Airbnb, founded in 2008 and public since 2020, carries a market capitalization of approximately $93 billion, so the scale of that deployment challenge is considerable.

The internal changes he described feed directly into the products guests see. Airbnb used an internal organizational context graph called Everest, which relies on technologies like LLMs, embeddings and AI-based retrieval, to build and query its graph. The grocery delivery service was built first, and the learnings from it were made available in Everest, which let the airport pickup team move much faster because the two services share a similar shape: API integrations with partner companies layered onto Airbnb's platform. Al-Dahle also said the context graph lets generalists work across very specialist parts of the codebase, which is a different answer to the handoff-heavy process of product requirements, design, engineering and testing that traditional software teams rely on.

The stakes appear to hinge less on whether the models are good enough than on how Airbnb keeps engineering judgement intact, especially among junior staff who now work alongside systems that produce so much of the output. Al-Dahle's answer is a rule that every engineer must be able to explain what was built, even when an AI generated the pull request; whether that rule holds as asynchronous agents spread from on-call triage to fraud, trust violations and marketplace quality is likely to be the thing worth watching.