PowerUp Coaching — Living Knowledge Base

Write Loops, Not Code — The Shift That's Changing How Experts Work With AI

3 min read

Topic

The shift in how skilled people direct AI: from writing step-by-step instructions to defining goal-seeking loops (goal + rubric + stop condition) and letting the model iterate — including loops nested within loops.

Target Reader

An intermediate AI user — coach, consultant, builder — who has moved past one-shot prompting and now runs multi-step work with AI, but still babysits each step manually and wonders why their outputs plateau.

The Fear / Frustration / Want / Aspiration

“I’m good at prompting, but I’m still hand-holding the AI through every step. The pros seem to get the AI to just… go and finish things. What’s the move I’m missing that turns AI from an assistant into something that completes work?”

Before State

The reader specifies how to do each step and checks every result. Their leverage is capped by their own attention — they’re the loop. Output quality plateaus because they’re optimizing instructions, not outcomes.

After State

The reader specifies what good looks like — a goal, a short rubric, a stop condition — and lets the model iterate to meet it. They nest loops for complex work (research loop → draft loop → headline loop) and use effort level as a persistence knob. Their attention moves from execution to defining success.

Narrative Arc

The Claude Code developer says he doesn’t write code anymore — he writes loops. That sounds like a coder’s quirk until you realize it’s the general shape of expert AI work now. The turn: the unit of work has shifted from the instruction to the loop. The resolution: a concrete method for converting a babysat task into a self-running loop, and for nesting loops on complex work.

Core Argument

The highest-leverage move with AI is no longer writing better instructions — it’s defining a goal, a rubric for “done,” and a stop condition, then letting the model loop until it meets them.

Key Evidence / Examples

  • “Boris Cherny — the Claude Code developer — says he doesn’t write code anymore. He writes loops.” — Lou
  • Loops within loops: a writing team loops on research, then drafts, then the headline — each until acceptable
  • Two control knobs: stop condition (a loop without one is a runaway) and effort (raises persistence, not intelligence)
  • “Not every problem deserves loops within loops at high effort” — calibration matters (Insight - Effort Can’t Buy Reasoning — The Model Sets the Ceiling, Effort Sets the Depth)

Proposed Structure (5–7 beats)

  1. The frontier-developer quote that reframes everything
  2. From instruction to loop — the unit of work has changed
  3. Anatomy of a loop — goal, rubric, stop condition
  4. Why stop conditions are non-negotiable (the runaway failure)
  5. Loops within loops — composing complex work
  6. The effort knob — persistence vs. intelligence
  7. Convert one task today — a worked example

Editorial Notes

Actionability scored a notch lower (3) because the payoff is conceptual — the “convert one task” exercise must carry the practical weight; make it vivid and specific or the article stays abstract. Pairs with the “Don’t Rent Autonomy” brief (this is the concept; that is one application). Verify the Boris Cherny attribution/quote before publishing.

Next Step

  • Approved for drafting
  • Needs revision
  • Deprioritised

Graph View