A plan-act-review loop splits agent work into controlled stages that repeat until the task is verified.
Plan. The agent inspects the task and proposes steps, assumptions and risks. It also lists the files it expects to change. A person or policy approves the direction.
Act. The agent performs one bounded step and gathers evidence, such as test output.
Review. The agent and supervisor compare the result with the plan and acceptance criteria. They correct mistakes, update assumptions and choose the next step.
This is safer than one long run because a wrong direction shows up early. Plans should stay light for small tasks. They become more detailed as impact, uncertainty and irreversibility grow.
Rewriting in plainer words…
This answer doesn't lend itself to a diagram - it reads best . No credits were charged.
Saved in this browser - sign in to keep your review list.
How should your speech become text?
Listening… your words appear above as you speak - tap Stop when you're done.
Recording · cr - tap Stop & transcribe when you're done.
Transcribing with AI…
Voice:
Keep going - a few more words and AI can grade it.
Interview lens
How interviewers actually use this question
Likely follow-ups
How do you decide how much planning a task needs before the agent starts acting?Scale it to impact, uncertainty and irreversibility. A one-line fix needs almost none. A data migration needs steps, risks and a way back.
What counts as evidence after an act step, and why isn't the agent's own summary enough?Test output, diffs, logs, running the feature. Agents can report work as done when it was skipped, so check artifacts, not claims.
A plan-act-review loop breaks an agent's work into three stages that repeat until the task is verified, instead of one long unsupervised run.
In the plan stage the agent reads the task and proposes steps, assumptions, risks and the files it expects to touch, and a person or a policy approves that direction.
Then it acts, one bounded step at a time, and collects evidence like test output instead of just saying it's done.
In review, the agent and whoever supervises it compare the result with the plan and the acceptance criteria, fix mistakes, update assumptions and pick the next step.
The payoff is that a wrong direction shows up after the first step, not at the end of a long run when it's expensive to unwind.
I'd keep plans light for small tasks and make them more detailed as impact, uncertainty and irreversibility go up, because a heavy plan for a typo fix is just overhead.
Weak answers to avoid
Plans once, then lets it run to the endThat's the long run the loop exists to prevent. Say that act is one bounded step and review follows each step, so drift gets caught early.
Review means the agent says it's doneAn agent's own report can be wrong or skip work. Say review compares evidence like test output against the plan and acceptance criteria.
Same heavy plan for every taskA detailed plan for a one-line fix wastes time. Say plan depth grows with impact, uncertainty and irreversibility.
Leaves out who approves the planIf no person or policy approves the direction, the plan is only a description. Name the approval step that comes before the agent acts.
Interview lens
Likely follow-ups, what you can say, and the weak answers to avoid.
