A Clean Run

A hand-written AI agent run, line by line. The lesson: think → act → observe → repeat.

task
What can I make with what's in my fridge?
Watch this run on the timeline →
plan
First, a plan. I think in my head too, but anything I don't write down is gone.
  1. Pick the dish
  2. See what's actually here
  3. Find a stand-in
  4. Make the stand-in
  5. Final check, then cook
your call
Dinner is your call, not mine. Decisions like this are yours, so I stop and ask.
Pick tonight's dinner:
“nashville hot chicken” or “honey butter pancakes”
The script writes both paths. Beats tagged with a pick happen only on that path.
if you picked “nashville hot chicken
readthe Nashville hot chicken recipe
Hot chicken it is. First, the recipe.
if you picked “honey butter pancakes
readthe honey butter pancakes recipe
Breakfast for dinner. First, the recipe.
if you picked “nashville hot chicken
the dip needs 2 cups buttermilk — there is no buttermilk
No buttermilk for the dip. I have no eyes — your kitchen only exists for me when a tool reports it.
if you picked “honey butter pancakes
the yeast batter needs 1¼ cups buttermilk — there is no buttermilk
No buttermilk for the batter. I have no eyes — your kitchen only exists for me when a tool reports it.
One missing ingredient doesn't cancel dinner. What's actually in there?
lookthe fridge and the pantry
check what's real before deciding anything
chicken quarters, eggs, milk, one lemon, flour, butter, honey, a packet of yeast, hot sauce, pickle juice, cayenne, smoked paprika, brown sugar
Chicken, eggs, milk, a lemon, hot sauce, cayenne.
Everything the recipe needs is here except the buttermilk. One thing to solve.
readbuttermilk stand-ins
confirm the swap instead of hoping
milk + a squeeze of lemon, 10 minutes — a real buttermilk stand-in
Yes — milk plus lemon juice makes buttermilk.
The acid does what buttermilk does. I was already sure. But a memory and a guess feel identical to me, and I sound just as confident either way. Answering from the guess is called a hallucination. That's why I looked.
dostir lemon juice into the milk, let it sit
Milk, lemon, ten minutes. That's the swap.
if you picked “nashville hot chicken
thickened — whisk in the eggs, hot sauce, pickle juice
The dip is ready. Not done yet.
if you picked “honey butter pancakes
thickened — ready to wake the yeast with
The stand-in is ready. Not done yet.
lookevery ingredient against the recipe, line by line
check the full recipe once more before cooking
every line covered — nothing missing
Everything's covered. I checked the whole recipe again instead of trusting my memory.
done
think → act → observe → repeat
  • ·I wrote the plan first — thoughts I don't write down are gone.
  • ·I never saw your kitchen — I only knew what each look reported.
  • ·I looked up the swap — answering from a guess is a hallucination.

Every line above is a hand-written script; no model.

The same lesson with code: A Clean Run — “Tests are failing in utils.test.ts — find out why and fix it.

the other runs:
  • A Plan Fails — “/api/orders is returning 500s — fix it.
  • Memory Fills Up — “Rename getUser → fetchUser across the codebase — 14 files.
  • A Plan Fails — “I plan eight hours of work and get four. Figure out why.
  • Memory Fills Up — “I saved 14 apartment listings. Which ones should I actually tour?