Strategy question
How would you decide which tasks are safe to fully automate vs. require human approval?
- Manus
- Strategy
- Hard
Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.
Start a mock interview on this question · Mock interview from a job description
What this question tests
Judgment framework for delegating autonomy, similar to the Devin routing question but generalized to any autonomous agent.
How to approach it
- Establish the deciding criteria: reversibility of the action, blast radius if wrong, and how verifiable the outcome is before it takes effect.
- Fully automate tasks that are reversible, low-blast-radius, and easily verifiable, such as research summarization or draft content generation.
- Require human approval for tasks that are irreversible or high-blast-radius, such as sending an email on the user's behalf, making a purchase, or deleting data.
- For ambiguous middle-ground tasks, default to requiring approval initially, then relax to full automation only after the agent has built a track record of accuracy on that task type for that specific user.
- Build a lightweight tagging system so task types carry a risk classification the system can act on consistently.
- Define success as automation rate rising over time within safe categories, while maintaining a strict zero-tolerance bar on unauthorized irreversible actions.
What a strong answer includes
- Uses generalizable criteria (reversibility, blast radius, verifiability) rather than a fixed list of specific tasks.
- Proposes a track-record-based graduation path from human approval to full automation, which is a realistic trust-building model.
- Names concrete irreversible-action examples (purchases, deletions, sending communications) that should always require approval.
- Balances increasing automation (efficiency) against a strict safety bar (zero unauthorized irreversible actions).
- Proposes a practical mechanism (risk tagging) to apply the framework consistently at scale.
Common mistakes
- Giving a static list of task types without an underlying principle that generalizes to new task types.
- Not addressing how trust is built or re-evaluated over time for ambiguous tasks.
Likely follow-up questions
- How would you handle a task that starts low-risk but becomes high-risk mid-way?
- Would trust levels be per-user or apply system-wide?
- How would you audit whether the automation was working safely?
More strategy questions
- How should a general-purpose agent like Manus differentiate from ChatGPT Agent and Comet?Manus · Strategy · Hard
- How would you price an autonomous general agent so cost aligns with value delivered?Manus · Strategy · Hard
- Google Keep is a free product to save, share notes etc. How would you make it a subscription product & monetize it?Google · Strategy · Hard
- How would you launch (roll out) Amazon Go?Amazon · Strategy · Hard
- With an unlimited network bandwidth what would you build?Google · Strategy · Hard
- Strategize and implement omni-channel initiatives to improve customer LTV for Walmart.Shopify · Strategy · Hard
More questions from Manus
Learn the skill behind it
Chapters of the AI PM course that teach what this question tests.
- Chapter 4: Discovery and strategy for AI products
- Chapter 9: Prove it paid off: outcomes, economics, and pricing
- Chapter 14: Get the job: the AI PM interview loop