Working with AI · Field notes
Get more out of every model.
Same model, same price — what changes the output is how you drive it. Eight habits I lean on to get more out of any LLM, the same ones behind the work on this site.
Signal gain
Out of the boxDriven well
8
habits, one playbook
8
Eight habits. One playbook.
Same models everyone else has — pointed to do more. The difference was never the model. It's how you drive it, and these are the habits that compound.
The playbook
Eight habits that compound.
Same model in, sharper output out — the moves behind the work on this site.
-
01
Talk. Don't type.
Dictation runs about 3× the words per minute typing does, and talking drags out the context, caveats and specifics a model needs. Explain it like you would to a teammate — the detail comes free.
The moveSpend 90 seconds out loud: where it breaks, what you've tried, and what "done" looks like.
-
02
Load a skill before you ask.
Point the model at a specific skill — front-end design, a marketing framework — and it executes a real discipline instead of a generic answer. No skill yet? Have it build one, then reuse that expertise forever.
The move"Use the frontend-design skill, then redesign this page."
-
03
Anchor to a goal.
Declare the end-state up front and every step stays aimed at it, instead of drifting message to message. Use the goal feature where you've got it — it keeps a long session pointed at the outcome you actually want.
The move/goal "ship the pricing page, live and verified, by Friday"
-
04
Crank the reasoning to max.
For anything genuinely hard, turn reasoning up to maximum and let the model think before it answers. The extra thinking time pays for itself — a right answer on the first pass instead of a fast one you have to redo.
The moveOn a gnarly problem, set effort to max and let it work the whole thing through.
-
05
Buy the seat, not the credits.
Consumer subscription tiers ship far more usage per dollar than metered API credits. For hands-on daily driving, a flat plan used hard goes further — save the API budget for what truly needs automating.
The moveRun everyday work on a flat consumer subscription; save metered API for automation.
-
06
Make it show its work.
Give the model eyes on the result — let it run the test, read the error, see the screenshot — and a tight loop to fix what it finds. It closes the gap itself in seconds instead of handing you a draft to debug.
The moveWire it to the real output: run, read, fix — then show you the green.
-
07
Front-load the truth.
The model works from what's in the window, so hand it the real file, the real error, the real constraints up front. Solid context beats clever phrasing every time.
The movePaste the actual file, stack, and the exact error — then ask.
-
08
Verify before you trust.
Confident and correct are different settings, so make the work prove itself — run it, hit the URL, show the output — before "done" counts. The best results arrive with their own evidence.
The moveAsk for proof: "200 OK — here's the live page rendering," then move on.
The same playbook, applied
These eight habits are how the work on this site gets built.
Same models everyone else has — pointed where they pay for themselves. Want that applied to your stack?