Airbnb’s ‘Inside-Out’ AI: Code, Support, and Product Delivery Accelerated by AI Agents
Airbnb now uses AI across its engineering, support, and operations, with around 60% of new code AI-generated, about half of support tickets resolved by AI, accelerated rollout for new services, and internal tools empowering generalist developers. Developers at Airbnb must still verify and understand AI-generated work, ensuring engineering craft is maintained.
What changed?
Airbnb has significantly restructured its engineering and operational processes around AI systems. Notable changes include:
- Around 60% of Airbnb’s new code is authored by AI, resulting in 80% more features delivered year-over-year and a 1.6x increase in average engineer pull-request throughput.
- Approximately half (45%-50%) of customer support tickets are resolved fully by AI systems, with higher-risk issues still handled by humans.
- AI-powered tooling (such as Everest, an internal context graph) enables generalist engineers to work efficiently across specialized codebases.
- Airbnb has rapidly launched new services (groceries and airport pickups) with AI accelerating development cycles—from 8-9 months (groceries) to as little as 6 weeks (airport pickups).
- The company operates with at least 10 production AI models, mixing both open- and frontier-models, with use-case-specific tuning and evaluation for cost, latency, and accuracy.
- Automation agents (AirChat) increasingly handle operational tasks like incident triage and on-call procedures.