Hot take: not always.
I do not think everyone should run agents overnight by default.
For small tasks, it can be wasteful:
- burns tokens
- creates noisy diffs
- explores wrong rabbit holes
- adds review overhead for little value
Where night runs actually help is long-cycle work:
- long refactors
- broad test generation
- dependency cleanup
- docs and migration drafts
This only works when the task has enough complexity to benefit from multiple long tries while you are offline.
In the morning, I do a strict review pass:
- keep what is useful
- discard noisy changes
- lock decisions that still need human judgment
My rule: Use overnight agents for complex, high-latency tasks. Skip them for straightforward work.
Agents are leverage, not a checkbox.