You want human oversight without reviewing every small step. What should you do?
Why C is the answer
Place checkpoints at high-risk actions, and let low-risk work flow
Risk-based checkpoints keep human oversight where it matters, without slowing delivery.
In detail
- Place checkpoints at high-risk actions.
- Low-risk work can flow, so people review what matters.
- Add a checkpoint after any incident that shows a gap.
Use case
An illustrative scenario, not a real incident.
A team reviews database migrations and permission changes, and it lets formatting flow. People spend time where mistakes cost most.
See it in different ways
flowchart TD
A{"How risky is the action?"} -->|"Low and reversible"| B["Agent acts, people review later"]
A -->|"Medium"| C["Agent proposes in a pull request"]
A -->|"High"| D["Human approval before acting"]sequenceDiagram
participant A as Agent
participant V as Validation
participant R as Repository
A->>V: Structured plan
alt Plan passes
V->>A: Go ahead
A->>R: Makes changes in a branch
else Plan fails
V-->>A: Fix the plan first
endflowchart LR S0["Low-risk edits"] --> S1["Checkpoint at migration"] --> S2["Low-risk edits"] --> S3["Checkpoint before merge"]
Key point: Put checkpoints at high-risk actions.
Why the other options are wrong
- A. Remove the oversight from everything that looks like a small change
- Small-looking changes can still be risky, so some oversight stays.
- B. Require a person to approve each file edit as the agent works
- Approving each edit is what you are trying to avoid.
- D. Run every agent task only outside of normal working hours
- Timing does not provide oversight.