# AI Coding Productivity Review Worksheet

Review an engineer’s AI tool costs, completed code, quality and delivery. Then decide whether to renew a subscription, raise a limit, remove a verified unused charge or change the workflow. Review the work with the engineer before making the recommendation.

## Set the scope

| Field | Notes |
| --- | --- |
| Engineer, reviewer and review date | |
| Baseline and follow-up dates | |
| Team, repositories and main tasks | |
| Context: incidents, leave, onboarding or dependencies | |
| Missing tool activity, account links or work outside connected systems | |

Keep dates and filters consistent across reports. Record open PRs at each period’s end and any change in responsibilities or task difficulty.

## Compare costs and completed work

In Weave, open [AI → Overview → Cost per person](https://docs.weaveos.com/reports/ai), inspect the usage rows, then open the [individual report](https://docs.weaveos.com/reports/individual) and selected contributions. Fill the baseline now and the follow-up after the agreed change.

| Measure | Baseline | Follow-up | Source or context |
| --- | --- | --- | --- |
| Paid subscription and overage charges | | | |
| Per-person usage cost and fidelity labels | | | |
| Contracted seats subtotal and prorated period | | | |
| AI-attributed code output | | | |
| AI cost per output | | | |
| Merged PRs and what they delivered | | | |
| Quality: relevant bugs, reverts or rework | | | |
| Delivery: relevant cycle time or review delays | | | |

Per-person **Total cost** sums usage rows and may show list-price consumption on a flat-rate plan. **Contracted seats** is a separate, day-prorated subtotal. Use **ROI → Cost by fidelity** for the full total including seats, then reconcile it with the bill. Keep consumption estimates separate from paid charges. [Cost definitions](https://docs.weaveos.com/metrics/ai-tool-costs)

[Code Output](https://docs.weaveos.com/metrics/code-output) estimates expert effort in hours; it does not record actual working time. Output is counted at PR merge, so spending and completed work can fall in different periods. AI cost per output divides AI cost by AI-attributed output.

**What explains the costs and results, after reviewing the PRs with the engineer:**


## Optional: inspect an AI session

When the PRs leave a question about failed attempts or rework, open a relevant scored session. [AI Skills v2](https://docs.weaveos.com/releases/ai-skills-v2) uses eligible sampled conversations and shows a rationale and conversation location. Read that moment with the engineer and record one useful finding. Check collection coverage before interpreting a missing score; the sample does not cover every interaction.

**Session link and finding:**


## Make the recommendation

Verify usage coverage and the engineer’s needs before removing a charge. Name the paid charge that would change, or record that the decision has no billing effect.

| Decision | Agreement |
| --- | --- |
| Renew, raise a limit, remove a verified unused charge or change the workflow | |
| Reason, with supporting PRs or cost records | |
| Paid charge affected, amount and billing period | |
| Owner and start date | |
| Expected result, quality check and follow-up date | |

## Review the result

Complete the follow-up column using the same cost basis and comparable work. Note changes in tools, task mix or coverage. Describe what changed without treating a before-and-after comparison as proof that the action caused it.

**Keep, adjust or reverse the decision, and why:**
