Strategy question

How would you decide which engineering tasks to route to Devin vs. a human?

Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.

What this question tests

Judgment on task allocation between autonomous AI and humans, based on task characteristics rather than blanket rules.

Define the decision criteria: task ambiguity, blast radius if wrong, and how verifiable the outcome is (does a clear test or spec exist).
Route well-specified, low-risk, verifiable tasks to Devin, such as writing unit tests, fixing well-reproduced bugs, or routine dependency upgrades.
Keep high-ambiguity or high-blast-radius tasks with humans, such as architecture decisions, security-critical changes, or anything touching production payment logic.
Build a lightweight triage step where a tech lead or the ticketing system tags tasks by these criteria before assignment.
Add a feedback loop: track which Devin-routed tasks needed heavy human rework and adjust the routing rules over time.
Define success as overall team throughput increasing without a rise in production incidents.

Uses concrete, generalizable criteria (ambiguity, blast radius, verifiability) rather than a fixed list of task types.
Gives specific examples on both sides, like unit tests for Devin versus payment-logic changes for humans.
Proposes a feedback loop that improves routing accuracy over time instead of a static one-time rule.
Balances the goal (throughput) with a guardrail (incident rate) to avoid over-delegating risky work.
Acknowledges this is ultimately a judgment call that should be revisited as trust in Devin's output grows.

Chapters of the AI PM course that teach what this question tests.