Steps

Review reliability and incidents

Summarize incidents, recurring failure modes, and recovery speed so you can target the biggest reliability bottleneck first.

Audit safety and governance drift

Check approval bypasses, access scope changes, and new compliance concerns to prevent silent control erosion.

Prioritize next-week actions

Commit to one reliability hardening task, one approval policy update, and one conversion-path improvement for the next sprint.

Why this workflow is useful

Teams often measure output volume and miss risk drift.

This weekly review gives a durable operating loop for post-launch quality.

Inputs (copy/paste)

Gather this evidence before the review starts:

Incident log for the current week.
Approval bypasses or policy exceptions recorded.
Conversion data for key routes (/business, /business/apply, /obsidian-ai-plugin).
Open reliability tasks from last week.

Copy/paste template

# Weekly AI Risk Review

## Week of
- Date range:
- Reviewer:

## Reliability
- Incidents this week:
- Top recurring failure mode:
- Mean time to recovery:

## Safety and Governance
- Approval bypasses detected:
- Access scope changes:
- New compliance concerns:

## Product and Revenue
- Workflows with highest ROI:
- Workflows with poor ROI:
- Biggest conversion blocker:

## Actions for next week
- [ ] Reliability hardening task
- [ ] Approval policy update
- [ ] Conversion-path improvement

Definition of done

This review is complete only when:

One reliability fix is assigned to an owner with a due date.
One governance update is approved and documented.
One conversion-path bottleneck is selected for next-week testing.

If you only fill the template but do not assign ownership, you are logging risk instead of reducing risk.

Failure modes + fixes

Failure mode: Incidents are listed without a recurring-pattern diagnosis. Fix: Name one top failure mode and why it keeps repeating.
Failure mode: Governance drift gets noted but not converted into policy changes. Fix: Ship one approval policy update every week this section has findings.
Failure mode: Conversion blockers stay vague. Fix: Attach each blocker to one route and one measurable signal.

What to do next

Roll findings directly into next sprint planning and keep the owner list visible.
If you want implementation support with governance, apply here: AgentOps application.

Related workflows