ai-velocity-batch-compile
AI-Velocity — Batch-Implement, Verify Once, Triage Errors in Parallel
Section titled “AI-Velocity — Batch-Implement, Verify Once, Triage Errors in Parallel”For any large or multi-file task, DO NOT use the human-paced edit → verify → wait → fix → edit loop one change at a time. Blind-implement the whole batch — write/edit every file up front — then verify ONCE at the end, collect the full error set, and fix it in one more batch. Two verification cycles for a 20-file refactor, not twenty.
Why: the verification round-trip runs at human-wait speed and is the dominant cost; per-edit gating throws that away and multiplies flake opportunities. Full rationale: docs/ai-velocity-batch-compile.md.
Reserve incremental verification for exploratory work where the API is unknown and one probe de-risks the rest; parallelize independent file groups per parallelize-batch-work.md / parallel-teammate-git-index-race.md.
Related
Section titled “Related”ai-driven-design.md · parallelize-batch-work.md · docs/ai-velocity-batch-compile.md (anti-patterns, mindset, engine gotchas, commit/push verify-gate, history)