When to reach for it
- The result can be iteratively improved.
- External validation triggers a retry.
- Tool results necessitate new iterations.
Loop.
Repeat until good enough, or budget runs out.
One or more steps are repeated until a specific budget, quality bound, or exit condition is reached.
When it backfires
- There is no stable exit condition.
- Costs can spiral out of control.
- Errors amplify through repetition.
The tradeoff
Adaptive improvement is achieved at the risk of endless or highly expensive execution loops.
A loop you can draw on a napkin.
A body step produces output; a check step decides whether to exit or loop again.
Two ways this pattern will hurt you.
No exit condition — the loop spins forever
The check step always returns 'retry' because the quality threshold is set impossibly high.
Fix · Combine quality gate with a hard iteration cap. Treat cap exhaustion as a normal outcome, not an error.
Errors amplify through repetition
Each loop appends to the same context. A bad intermediate gets baked in and the model compounds it rather than fixing it.
Fix · Keep only the latest iteration in context, or use a scratchpad that the model must explicitly update.