Airbnb CTO: 60% of code AI-written, AI resolves half of support tickets
Latent Space · rss · 2026-10-02
Latent Space interviews Airbnb CTO Ahmad Al-Dahle (ex-Meta genAI head who led Llama launches) on making Airbnb "AI-native" via an inside-out approach: transform internal development first, then the guest experience.
Engineering gains
- 60% of code is AI-authored; 80% more features shipped YoY; 1.6x PR throughput per engineer
- Product/design/engineering now collaborate directly on prototypes instead of handing off docs — "code is the artifact"
- Internal agent AirChat carries org-wide MCP context; teams are starting to automate on-call with async containerized agents that triage alerts, open PRs, and close incidents
Customer support
- Roughly half of tickets are resolved purely by AI (Q2 earnings said 45%), with deliberate exclusions for high-stakes cases like safety issues
- Heavy synthetic-data testing before production made the rollout safe
Model strategy
- A "multi-model company" mixing frontier and open models, doing most post-training/RL on open models, with 10+ customized models in production
- Evals per use case along a cost-performance-latency Pareto frontier: strongest frontier models for coding (defects are costly), small post-trained models for latency-sensitive search — sometimes beating frontier on narrow tasks
Everest context graph
- An internal LLM/embedding-powered org context graph: grocery delivery took 9 months, airport pickup reused its learnings and shipped in 6 weeks; also lets generalist engineers work across specialized codebases
More from coding & agent
- New 354-page LLM practical guide ships with 250+ Python snippets, agent chapters — RexDouglass · 2026-10-02
- Aviation's ASD-STE100 controlled language as an anti-AI-slop prompt hack, and where it fails — Paimaamu · 2026-10-02
- Pi Durable as statecharts: an interactive demo of crash-safe LLM agent harnesses — sloppenheimer · 2026-10-02
- Microsoft open-sources NVX, an ultra-light OpenVMM-based micro-VM sandbox for agentic workloads — unixterminal · 2026-10-02
- exe.dev's 'Run Fewer Agents': why task management isn't the fix for agent sprawl — charles_irl · 2026-10-02
- Building an agentic ML team: multi-agent pipeline with 40% token savings — kmeanskaran · 2026-10-02