Field Guide · Using Kate

How to Actually Talk to an AI
and get the answer you needed

Most people get mediocre answers from AI because they ask mediocre questions. The model isn't the limit — the prompt is. Here's how to brief an AI like Kate so it gives you the pro-level answer instead of the safe, generic one. Works for a no-start diagnosis or a business plan.

Start here

Give it what a good tech would ask for

Imagine handing the job to the sharpest person you know — but they can only see what you type. What would they need before they could actually help?

  • Context — the vehicle, the board, the symptom, and what you already tried. Not "it won't start." The whole picture.
  • The real goal — are you after the root cause, the next test, a second opinion, or wording for a customer? Say which.
  • Your constraints — tools you have, time, budget, what you've already ruled out.
The rule: the AI mirrors your precision. A sloppy one-line question gets a hedged, cover-all-bases answer. A sharp brief gets a sharp answer.
Step 1The four things every good prompt has
  • ContextWho/what/the situation — "2014 Silverado 5.3, crank no-start, spark + injector pulse present"
  • GoalWhat you actually want — "narrow it down" vs "explain how this system works"
  • ConstraintsTools, time, ruled-outs — "I have a scope and a scan tool; fuel pressure is good"
  • FormatHow you want it back — "ranked list with the test for each," "one paragraph," "step by step"
Step 2Prompt patterns that pull better answers

Give it the whole case

Context wins
  • Dump the details a good diagnostician would want — symptoms, freeze-frame, what changed, what you ruled out.
  • More relevant context = a narrower, more confident answer. You can't over-inform it.
Weak"Why won't my truck start?"
Strong"2014 Silverado 5.3, crank no-start after sitting overnight. Spark and injector pulse present, 55 psi fuel. Starts on a shot of ether. Where would you look next?"

Ask for the reasoning

Show the work
  • Add "walk me through why" — you'll catch a wrong assumption before you chase it.
  • The reasoning is often more useful than the verdict, and it lets you correct the AI mid-thought.

Constrain the output

Shape it
  • "Give me the 3 most likely, ranked, with the one test that confirms each."
  • Telling it the shape you want turns a wall of text into something you can act on at the bench.

Keep the thread going

It's a conversation
  • Don't restart. Feed it the result of the test it suggested — it refines from there.
  • Correct it when it's wrong. "No, injector pulse IS present" and it re-routes. That back-and-forth is where the good stuff lives.

Ask it to challenge you

The pro move
  • "What am I missing? What would rule this out?" flips it from agreeing with you to pressure-testing you.
  • Best way to catch your own tunnel vision before you throw a part at it.
What gets you a bad answer

The habits that waste the tool

  • One vague line. "Best way to fix a P0300?" — no vehicle, no context. You'll get a textbook, not a diagnosis.
  • Hiding what you tried. If you don't say what you ruled out, it'll send you back over old ground.
  • Yes/no when you want depth. Ask "is it the coil?" and you get a coin flip. Ask it to reason and you get a path.
  • Taking the first answer as final. The first reply is a starting point. Push, narrow, feed it data.
  • Not naming the platform. "The module won't program" means nothing without the make, tool, and job.
HabitsHow the best users talk to Kate
  1. Lead with context, alwaysVehicle/board, symptom, what you tried — before the question. It sets everything that follows.
  2. Ask for ranked options + the test for eachYou want something you can act on, not a list of everything that could theoretically be wrong.
  3. Make it show its reasoning"Walk me through it" catches bad assumptions and teaches you the system at the same time.
  4. Keep one thread, feed it resultsDon't start over each time. The conversation gets smarter as you add data.
  5. Tell it when it's wrongCorrecting the AI isn't rude — it's how you steer it to the right answer fast.
Try it right now

Kate holds the whole case

The better you brief her, the deeper she goes — and unlike most AI, she won't dead-end you with "consult a professional." Give her a real one: the vehicle, the symptom, what you've tried, and what you want. Watch the difference a good prompt makes.