AI Coding Tip 040 - Ask the AI to List Where Your Prompts and Skills Contradict Each Other

Your AI coding tip for the day is to ask the AI to list where your prompts, skills, and decisions contradict each other before it starts making up a tie-breaker. This is important because when multiple instructions disagree, the AI often picks one at random and carries on confidently, creating a hallucinated rule that isn't mentioned in the output.

This can lead to your decisions drifting as you learn the domain, with old ones mixed in with new ones, creating hidden assumptions the AI makes for you. To avoid this, commit your work so the audit starts from a clean checkpoint, gather all the files the AI reads, and ask it to audit only, without making any edits or creating new features.

Request a numbered table of contradictions, quoting both conflicting instructions and naming the file or message where each one appears. Ask the AI to rank each item by what would break first if it guessed wrong, and also ask for a second list of decisions you never made explicit, surprising you with ones you may have forgotten.

Read both quoted lines of every item yourself, as the AI might invent contradictions on its own. After picking the winner of each item, write down the reason next to it and keep them in one place, replacing every other copy with a pointer in a pitfalls file for the traps you found. Run the same audit with a different model, as it may miss some blind spots.

Repeat this after every long session, as iterating helps you learn the domain and find new contradictions. By doing this, you can surface silent choices, find forgotten decisions, remove the tie-breaker, shrink your rules, learn the domain faster, and make decisions more durable. This simple prompt can save you a lot of time and trouble by avoiding hallucinations and helping you understand your own setup better.