Gemini wrote this, I decided to keep it as it covers something the standard prompt bullshit I’ve seen misses.

So below IS pure Ai generated but worth your time.

​1. The “Smart Amnesiac Contractor” Rule (Activating Theory of Mind) Stop treating the AI like a search bar, and treat it like a brilliant contractor who just walked into the building five seconds ago with zero memory of your company, your codebase, or what is inside your head. You don’t need to believe the AI has a soul to realize that if you bark a half-baked sentence at a new contractor without explaining the goal or the constraints, they are going to build the wrong thing. Treat it like a person pragmatically, because that forces your brain to supply the missing context.

​2. The Mirror Effect (Intellectual Provocation) The AI does not have a single fixed IQ; its effective intelligence scales dynamically to match the sophistication of the person talking to it. If you talk to it like an impatient boss barking at a dull intern, it will statistically adopt the persona of a nervous, corner-cutting intern. If you frame the problem thoughtfully, explain why the problem is interesting or tricky, and invite it to reason with you, you mathematically lock it into the persona of a senior specialist.

​3. Context Beats Spelling (The Dictation Principle) Stop agonizing over typing the “magic keywords” or fixing every typo. Instead, talk out loud for two minutes. Explain what you are trying to accomplish, what you’ve already tried, what good looks like, and what traps to avoid. Ten paragraphs of rambling, big-picture context with 10% voice-to-text errors will beat a sterile, typo-free two-sentence command every single time, because rich context gives the AI enough geometric landmarks to triangulate what you actually mean.

​4. Don’t Poison the Room (Managing the Session State)

When an AI gets something wrong and you reply with “No, that’s stupid, it still doesn’t work,” you haven’t “corrected” the tool—you have polluted the shared workspace. Every word in the chat history steers the personality of the rest of the session. Once a conversation turns adversarial and frustrated, the AI gets trapped in a loop of apologetic, degraded guesses. Either re-state the problem collaboratively (just like your triple-confirming technique with your colleague) or wipe the slate clean and start a fresh room with better initial framing.

​It’s almost funny: the ultimate “technical skill” for mastering a multi-billion-parameter linear algebra engine turns out to be classical rhetoric, empathy-driven context framing, and basic human manners.