ManualMode field notes
Keep the judgment AI makes easy to skip.
Practical playbooks for reading, testing, debugging, and owning AI-assisted code. Written for engineers, not search bots.
One operating loop
- 01Read the contract before the patch.
- 02Turn a suspicion into test evidence.
- 03Make the smallest safe correction.
Try it yourself
Practice reviewing AI-generated code for free.
Review two patches in your language. No account or code execution required.
Start here
Useful, original guidance.
New material ships only when product evidence or repeated engineer questions justify it.
01Refactor reviewCharacterization tests for AI refactoring: catch hidden mutationCompare an AI refactor with existing behavior, catch an in-place sort that changes the caller's array, and verify a minimal correction with Node.js tests.8 min02Retry verificationTest retries in AI-generated code: detect duplicate writesRun a local lost-reply simulation, observe a duplicate write, and test same-key replay plus conflicting input. Learn what the in-memory check cannot prove.8 min03API verificationContract tests for AI-generated API code: check the real JSON responseCatch a nullable field disappearing across a real HTTP and JSON boundary, then correct it and verify the response.9 min04Testing the testsMutation testing AI-generated tests: prove they catch a wrong answerRun a local shipping-boundary example, expose a surviving mutation, and add an independent assertion that rejects the bug.8 min05Project ModeHow to practice manual coding in your own projectsReserve a bounded stale-search fix, prove it with a runnable local regression test, and keep raw source local by default.8 min06Code ownershipVibe coding without losing code ownershipUse coding agents for speed while keeping the contract, verification, failure recovery, and rollback decisions in human hands.7 min07Verification playbookAI code review checklist: verify generated code before it shipsWork through a runnable JavaScript async-filter bug, then review contracts, edge cases, tests, and failure paths.8 min08Testing playbookHow to test AI-generated code before mergingDerive tests from the contract, attack plausible failure modes, review generated tests, and leave evidence another engineer can audit.7 min09Coding fitnessHow to code without AI when your skills feel rustyRun one 30-minute manual coding rep, verify it with tests, and use the evidence to choose what to practice next.7 min
Turn reading into evidence