Agentic coding at Airbnb

Presented by Szczepan Faber and Mike Nakhimovich (Airbnb) at DPE Summit 2025, an event developed and hosted by Gradle. Visit https://gradle.com/solutions/generative-ai/?utm_campaign=dpe-summit-2025&utm_medium=social-organic&utm_source=youtube&utm_content=agentic-coding-airbnb-main to learn how to reap the rewards of AI in today's software development landscape while avoiding the pitfalls. It’s amazing how we can now build working apps just by few-shot prompting #LLMs. But try doing this with #monorepos of 10s of MLOC, like the ones used for planet-scale apps that must be secure and compliant. Responsible #agenticcoding at scale, where 1000s of engineers materialize code changes by engaging in deep sessions with AI is really challenging. We want to share Airbnb’s journey, the trade-offs we picked, learnings, outcomes and productivity impact we observed. 00:00:00 Introduction to agentic coding at Airbnb 00:00:16 Bold vision for productivity growth in 2025 00:01:41 Paradigm shift to human steering multiple agentic sessions 00:01:51 Using pair programming analogy for agentic coding 00:04:02 Holistic methodology for measuring success and developer productivity 00:05:01 Rapid adoption curve of agentic coding tools 00:07:00 Defining the inner and agentic developer loops 00:08:25 Model for learning skills in agentic coding 00:09:53 Importance of human review before PR submission 00:12:22 Leveraging sandboxed cloud environments for parallel sessions 00:14:17 Meeting developers where they are across different tools 00:15:13 Evolution from IDE plugins to agentic CLI tools 00:16:39 Analyzing IDE-first versus CLI-first development strategies 00:19:14 Formation of a core internal Devi team 00:19:53 Three-part bet: agents, Airbnb awareness, and MCP standard 00:21:27 Developing Air Chat abstraction for deploying multiple engines 00:22:09 Early adoption and investment in the MCP standard 00:24:00 Building the MCP ecosystem with off-pave paths 00:26:01 Structuring orchestrators and agents using Anthropic’s patterns 00:27:32 Philosophy of not reinventing wheels and using open source 00:28:44 Integrating Airbnb-specific knowledge into agentic search 00:30:33 Final lessons: community empowerment, barrier removal, and maintaining standards #AgenticCoding #AIAgents #DevProductivity #AirbnbTech #MCP #ToolCalling #IDEs #CLI #IntelliJ #VSCode #CodeReview #PRVelocity #RAG #Monorepo #DevEx #EngineeringLeadership #DevTools #SoftwareEngineering