Purpose

The first formal review after an agent enters production at M15. Scheduled before launch (per template 07 pilot-to-prod checklist item 10) and held 30 days later.

Different from the quarterly review (template 15) in scope and depth:

• 30-day review: This agent only. Validation that the production deployment is healthy and the agent is delivering its promised KPI at production scale.
• Quarterly review: Portfolio-level. ROI, risk re-classification, framework changelog.

This review is also the first opportunity to consider autonomy promotion (Stage 1 → Stage 2) if performance supports it. Most agents stay at Stage 1.

• When you use it: Exactly 30 days after production launch (M15). Single agent.
• Who fills it: CoE Lead + Department Champion co-drive. Builder provides data.
• Time: 60–90 minutes meeting + 30 minutes write-up.
• Output: Signed review doc attached to the registry. Decision: continue / iterate / pause / promote autonomy / retire early.

Worked example (AP Accountant invoice reconciliation)

30-Day Review — finance-invoice-recon v1.0

Launch date: 2026-06-13 Review date: 2026-07-13 (30 days) Attendees: Mike Chen (Finance Champion), Morteza Moradi (CoE Lead), Sarah Patel (AP accountant, pilot user), Pat Lee (CISO — informed observer) Author: Morteza Moradi

1. KPI vs target (30-day actuals)

Headline: Agent meeting or exceeding every Card §12 threshold at 30 days.

2. Volume + adoption

Adoption note: Sarah reports she's used the time to start vendor-terms renegotiation work — a higher-value activity. This was the explicit reallocation goal from the intake form §5.

3. Incident summary

All Sev-3 incidents resolved by the runbook automatically. No post-mortems triggered.

4. Five monitoring signals — health check

5. Sarah's qualitative feedback (pilot user)

Direct quotes from her debrief:

• "It saved my mornings. I used to do invoices before my first coffee."
• "The confidence scores are useful — when it's < 80%, I look harder and usually it's the multi-line POs."
• "The cosmetic capitalization thing is annoying but not blocking — vendor name 'apple inc' vs 'Apple Inc' shows up in the draft email."
• "I trust it. But I still want the click-to-approve step. Don't take that away."

Sarah's overall: keep it as-is, fix the cosmetic in v1.1, don't promote autonomy.

6. Autonomy promotion consideration

7. Open items / v1.1 backlog

8. Framework feedback

Lessons from this 30-day review fed back to the framework:

• The 5 monitoring signals (framework §21) are working as designed — caught nothing because nothing went wrong. Worth noting in the framework changelog that "first 30 days clean" is a real possibility, not just a goal.
• The autonomy progression criteria in framework §22 should be supplemented with "user preference" — if the human user prefers Stage 1, that's a legitimate reason to defer promotion. Logged for framework v1.1.

9. Decision

✅ Continue at Stage 1. No major changes. v1.1 backlog confirmed (items 1–2). Next review: quarterly (Q3 2026, ~2026-09-13).

Sign-off

Blank template (copy below for your agent)

# 30-Day Review — [Agent ID] v[X.X]

**Launch date:** [YYYY-MM-DD]
**Review date:** [YYYY-MM-DD]
**Attendees:** [List names + roles]
**Author:** [CoE Lead or Champion]

---

## 1. KPI vs target (30-day actuals)

| Metric | Card §12 target | 30-day actual | Status |
|---|---|---|---|
| | | | |

**Headline:** [One sentence — meeting / missing / mixed.]

## 2. Volume + adoption

| Field | Value |
|---|---|
| [Volume metric] | |
| Unique users | |
| Total time / value saved | |

[Adoption note: any qualitative observations]

## 3. Incident summary

| Severity | Count | Detail |
|---|---|---|
| Sev-1 | | |
| Sev-2 | | |
| Sev-3 | | |
| Sev-4 | | |

[Post-mortem links if any]

## 4. Five monitoring signals — health check

| Signal | Status | Notes |
|---|---|---|
| Output distribution shift | | |
| HITL escalation rate | | |
| Decision audit trails | | |
| Cost per execution (step level) | | |
| Exception routing | | |

## 5. [Pilot user]'s qualitative feedback

[Direct quotes or paraphrased — what does the human user actually think?]

## 6. Autonomy promotion consideration

| Question | Answer |
|---|---|
| Days at current stage | |
| HITL acceptance rate | |
| Threshold for promotion (framework §22) | |
| Eligible for promotion? | |
| **Decision** | |

## 7. Open items / v1.X backlog

| # | Item | Severity | Owner | Target |
|---|---|---|---|---|
| | | | | |

## 8. Framework feedback

[Anything this review revealed about the framework itself — lessons, gaps, refinements]

## 9. Decision

[✅ Continue as-is / ⚠️ Iterate (v1.1 specified) / ⏸ Pause for investigation / ⬆️ Promote autonomy / ⛔ Retire early]

Next review: [Quarterly date]

### Sign-off

| Role | Name | Date |
|---|---|---|
| AI CoE Lead | | |
| Department Champion | | |
| User (informed) | | |

Usage notes

• Schedule before launch. Pilot-to-prod gate (template 07) item that the 30-day review is on the calendar. Don't let launch happen without it.
• The user's qualitative feedback (Section 5) often matters more than the metrics. A statistically successful agent that the user resents is failing.
• Autonomy promotion is rare. Most agents stay at Stage 1. Even when metrics support promotion, the user's preference is a valid reason to defer.
• Don't skip Section 8 (framework feedback). Every review is an opportunity to refine the framework. Lessons compound.
• 30 days is not the end of validation. Quarterly review (template 15) is the next checkpoint. This is just the first.

Common pitfalls

Framework cross-references

• framework.md §11.2 (per-agent lifecycle — 30-day post-launch check)
• framework.md §21 (5 monitoring signals — first formal check)
• framework.md §22 (autonomy progression — first promotion consideration)
• framework.md §29 (ROI tracking — 30-day data point)
• framework.md §22.1 EU AI Act Article 72 (post-market monitoring starts here)
• workflows.md Step 13 (Production launch — schedule the 30-day review)
• workflows.html → In Action view → node M15 → M16 transition (post-launch monitoring)