We built AI to split the monolith (engineering.monday.com)

monday.com announced Morphex, an AI-driven migration system that tore into their decade-old JavaScript client monolith and accelerated an effort once estimated at 8 person-years into a six-month program. Morphex automatically scans the entire client codebase, maps every file to items on a monday.com board (scored by dependency and risk signals), and drives a Research→Plan→Review pipeline orchestrated in Node.js. The system combines LLMs with deterministic codemods, static analysis, linters, tests, feature flags, and human-review hooks: small, verifiable steps, retry logic on failures, and “activate a human” todos whenever AI judgment is required. Results included automated TypeScript conversions, migration from Redux to Zustand, comprehensive tests, and PRs announced to engineers — Morphex extracted 1% of the client codebase in a single day. For the AI/ML community this is a concrete blueprint for scaling large engineering refactors: pure LLM prompting struggled with hallucinations and drift, but pairing models with strict orchestration, validation loops, and tooling made large-scale, parallel, autonomous migrations practical and safe. Key implications include treating AI as a managed workforce rather than a black box, using codemods/static analysis to supply reliable context to prompts, and designing systems that balance automation with human oversight for final approval — a repeatable pattern for other massive codebase rewrites and long-lived engineering automation.