Verify an AI coding suggestion before counting it as learning
At a glance
Luis Botelho uses a hypothesis-and-verification log to turn AI-assisted implementation into explainable engineering practice. This workflow is listed under Learning, Coding, Verification.
- Agents / tools
- ChatGPT, NotebookLM, Git
SourceBuilding a tool for these workflows? See weekly rates.
What they did
1) In a project with a working test environment, prepare ChatGPT for tutoring and NotebookLM for course sources and session notes. Create docs/ai-log.md with sections for the problem, your prior hypothesis, AI suggestion, errors, verification and a memory check. 2) Inspect the existing system and record your hypothesis before asking a bounded question. Give ChatGPT your attempted explanation or code and request criticism or a small exercise addressing the gap. 3) Compare its answer with documentation, implement or revise the code and test the intended behavior. Record incomplete suggestions, consulted documentation, experiments and results. 4) Close the chat and explain the solution from memory. Commit after verification and the recall check, and save where you stopped for the next session.
Worth knowing
The author's learning criterion is being able to explain, reproduce and verify the reasoning, rather than merely generating code.
