PATIENCE _
Leaving Better Notes
You write a system prompt. The agent runs for 47 steps. Step 23 goes sideways. Your instinct: interrupt, override, rewrite mid-run.
What worked: adjust the prompt and run it again.
You explain the architecture to your team. They nod. Two weeks later the PR looks nothing like what you described. Your instinct: take the keyboard, do it yourself.
What worked: ask better questions in the next review.
You read the book. You don't get it. You read it again. Still don't get it. Your instinct: force understanding, highlight harder, take notes on notes.
What worked: put it down. Come back in three months.
A man with no short-term memory hunts for the man who attacked his wife. He can't form new memories .. so he tattoos instructions on his body, takes polaroids, leaves notes. Every morning he wakes up as a stranger to himself.
Leonard tattoos instructions on his body because his future self won't remember writing them.
He's not controlling his future self. He's influencing a stranger who happens to live in his body.
You're always writing notes to someone who won't remember the context in which you wrote them.
The agent taught me patience with humans.
Because when the agent goes off-track on step 23, you can't reach into the context window and grab the wheel. There is no wheel. There's only: what you said at the beginning, and what it did with it.
Once you internalize that .. that influence is setup, not intervention .. you start seeing it everywhere.
The reason your teammate didn't build what you described isn't because they didn't listen. It's because your prompt wasn't clear enough.
The reason you didn't learn the thing isn't because you're slow. It's because you didn't give yourself enough runs.
re-run
questions
later
I debug my system prompts the same way I debug my management. The same way I debug myself.
Write the clearest note you can.
Let the run happen.
Observe what actually happened, not what you wanted to happen.
Adjust the note. Not the listener.
Run it again.
This is the practice I keep returning to: write a clear note, make the next action observable, and wait long enough to see what happened. The timing and feedback differ across agents, teams, and personal habits.
Leonard leaves notes for the next version of himself.
A prompt, a culture doc, and a habit do something smaller:
they give the next attempt a place to start.
The note cannot control the next reader.
It can make the next step clearer.
