Agents aren't junior engineers: a 12-rule discipline for unattended Claude Code runs
FinanceYF5 · x · 2026-11-12
A practitioner's rules for multi-agent coding: delegate tasks but not judgment (product tradeoffs, security boundaries and ship calls stay human); plan before code with Plan Mode, split parallel work into non-overlapping parts; answer five questions before unattended runs (definition of done, self-verification, scope, time/cost caps, who approves); restart after two failed corrections by rewriting the prompt in a fresh session; accumulate corrections into CLAUDE.md or shared skills; express preferences in prompts and hard rules in hooks; measure cycle time, rework and stability instead of PR count; scale verification (tests, reviewers) rather than attention; and remember AI makes building cheap, not maintenance.
More from coding & agent
- Engineers Using LLMs Should Understand Tokens, Context Windows and Hallucinations — _jaydeepkarale · 2026-11-12
- Building a Remote MCP Server for Real SaaS: OAuth 2.1, 70 Task-Shaped Tools — djskrilled · 2026-11-12
- Dev shows e-ink desk display tracking 20 Codex agent threads in real time — Dimillian · 2026-11-12
- Building a verifiable local-AI software factory on a single GPU: the judge can't be the defendant — d3nnyvg3org3 · 2026-11-12
- GraphRAG for CTOs: three retrieval patterns and a four-week test to decide — JeremyCMorgan · 2026-11-12
- Steve Yegge burns 5B tokens a day: agents first break everything around the model — FinanceYF5 · 2026-11-12