42-Hour AI Port: Code Generation Took 12 Hours, the Rest Was Human Review
martinwoodward · x · 2026-09-17
GitHub's martinwoodward breaks down a 42-hour large-scale code migration: nearly all code generation happened in the first 12 hours; the rest was validation and review. Agents compared old vs new behavior while humans judged architecture and risk — "writing the code is the start, not the finish." Follow-ups: 15 agents built/tested on one laptop, with Stephen Toub using a chat session as a build scheduler to serialize builds over shared resources.
Related event: 42-Hour AI Migration: Code Generation Took Only 12 Hours(2 posts)→
More from coding & agent
- The harness tax: Claude Code costs 2x Pi at the same 15.3 turns, with 10x initial context on SWE-bench Lite — HamelHusain · 2026-09-17
- Jev: an open-source action-picker that splits agent thinking from clicking — alexcovo_eth · 2026-09-17
- CROA open-sources a deterministic execution layer enforcing trajectory-level constraints on AI agents — CROA_PROJECT · 2026-09-17
- Moving coding agent memory off-machine: from Dropbox CONTEXT.md to MCP shared memory — Asly97 · 2026-09-17
- What's the worst 'fix' you've seen an AI coding agent make? — fromkrish · 2026-09-17
- Self-taught builder shipped 900 PRs in 3 months running 18 terminal agents to automate insurance brokerage — CShorten30 · 2026-09-17