AI-accelerated migrations at scale
summary
Airbnb reports a concrete migration playbook: a custom AI migration framework, idempotent validate-transform loop run concurrently (concurrency up to 100), and a sample tune sweep process. An enzyme migration of 3,400 files drops from an estimated year and a half to six developer weeks. The toolkit moves over a million lines across 12 unique migrations, uses scripts + CLI + pmap-style parallelism, and centers on the tenets Complete to 100% and Be a host while surfacing review bottlenecks and tuning prompts.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 0:29Enzyme migration challenge
- 2:13Year-long to six weeks
- 3:17Four techniques for AI migrations
- 4:01Validating with automated scripts
- 5:26Validate-transform loop
- 6:35Brute force retries
- 7:39Concurrent execution
- 8:44Sample-tune-sweep method
- 10:58In-house AI framework
- 12:47Key principles
- 13:30Q&A
related talks
28mAI Migration at SCALE: Make Complex Code Tasks 10X Faster (The Blueprint)
35mPerforming Infrastructure Migrations at Scale
26mAI and APIs: Ensuring Platform Migration Reliability with AI
30mAI First Legacy Migration
48mScaling a Migration to CD at Airbnb
25mMass migrations to the cloud
30mWhy most technology migrations fail (and what we've learned along the way)
26mA Pragmatic Journey from Monolith to Microservices Powered by AI:GenAI