Principle
Ask a coding agent for one coherent increment at a time
Autonomy works better when the unit of progress is small enough to verify.
When it fits
- An agent keeps touching many features, leaving partial implementations or declaring victory early.
When to avoid it
- Do not fragment a change so aggressively that every increment is unusable or creates temporary architecture debt.
Why it matters
Choose one ledger item or coherent slice, let the agent implement and test it, then persist the result before moving on. The slice can include multiple files, but it should end in an inspectable state. This reduces half-finished branches and makes failures easier to localize without forcing the human to micromanage line by line.
An example
Instead of 'modernize the whole integration layer,' first complete one endpoint migration including tests, logging and documentation.
Check your result
Each autonomous work interval ends with a coherent artifact, a verified status and a clear next item.
Keep this limit in mind
- Do not fragment a change so aggressively that every increment is unusable or creates temporary architecture debt.
Evidence and sources
Anthropic's long-running harness improved coherence by asking the coding agent to make incremental progress on one feature rather than attempting the entire application at once.
Some refactors are inherently cross-cutting; the increment still needs a coherent acceptance boundary.
Effective harnesses for long-running agents · Choose a single feature and leave progress artifacts