AI coding productivity · Sources checked
How to Measure AI Coding Productivity per Engineer
Compare AI spending with the code each engineer ships, its quality and delivery time. Use the results to review subscriptions and usage limits.
On this page
To measure AI coding productivity per engineer, compare AI tool costs with completed code, code quality and delivery time. Then inspect the work behind the numbers. This gives you a basis for deciding which subscriptions to renew, whose usage limit to increase and where training or a workflow change would help.
Weave connects AI tool usage and costs with code analysis for individual engineers. For connected tools such as Cursor, Claude Code and GitHub Copilot, you can compare spending with AI-attributed code output, open the underlying pull requests and check whether the work needed rework. Its individual reports bring those measures together so you can explain a budget decision with specific work.
Start with each engineer’s AI costs
In Weave’s AI report, Cost per person shows each engineer’s tool usage cost and cost per output. Open a person to see their tool mix, plans and cost details, including which amounts are billed and which are estimated.
Keep usage estimates and paid charges separate. With a flat-rate subscription, the per-person Total cost can show what the usage would cost at list prices, even though the company pays a fixed seat fee. Contracted seats is a separate subtotal. ROI → Cost by fidelity shows the full total including seats, broken down by how the cost is sourced. Reconcile that total with the provider’s bill before reporting savings. See the cost definitions.
This matters when reviewing a renewal. Reducing usage within an included allowance may leave the bill unchanged. Removing an unused paid seat or reducing a billed overage affects an actual charge.
Compare AI spending with completed code
A useful efficiency measure is AI cost per output: AI cost divided by AI-attributed Code Output.
Weave’s Code Output estimates the expert engineering effort represented by a change. It lets a difficult migration and a routine generated update carry different weights, even when each counts as one PR. The unit is estimated expert hours, rather than recorded working time or hours saved.
Open Individuals → AI to compare each engineer’s AI cost per output, AI cost and AI output share. The table puts the highest cost per output first. Open a person’s individual report to inspect their merged PRs and recent contributions.
Use consistent date ranges, repositories and cost definitions. Output is counted when a PR merges, so current spend may fund work that will appear in a later period’s output. Check open PRs before concluding that higher spending produced little work.
Compare similar assignments, or follow the same engineer across comparable periods. A migration, an incident and a routine feature can require very different amounts of investigation. The PRs explain differences that a ratio alone cannot.
Check code quality and delivery time
Review the quality of the code alongside the amount delivered. Weave’s Individuals → Quality view includes code quality, bugs per output, reverted PRs and code turnover. Open the relevant contributions to see what changed and what needed correcting.
Then check Process and Review for cycle time, task lead time and review turnaround. If an engineer finishes a change quickly but it waits in a review queue, increasing their AI allowance will not address that delay. If output rises alongside avoidable rework, inspect the code and verification process before expanding the same approach.
Together, these checks explain whether the problem is the cost of producing code, the quality of that code or a delay in getting it reviewed and delivered.
Use AI sessions to explain repeated problems
When the PRs leave a question unanswered, inspect a relevant coding conversation. Weave’s AI Skills v2 assesses how engineers plan, provide context, verify results and direct AI tools. Each assessment includes its reasoning and a location in the conversation.
For example, repeated failed attempts may point to missing requirements. A change that needs rework may reveal a verification step that was skipped. The conversation gives the engineer and manager something specific to improve.
AI Skills uses eligible sampled sessions, so check collection coverage before interpreting a missing assessment. The release documentation explains the sampling rules and supported collection methods.
Intellistack used Weave to compare AI adoption and output across components. Backend teams were benefiting more than frontend teams, which gave leadership a training priority. After training on Cursor plan mode and automated test writing, leadership tracked whether engineers adopted those practices and whether adoption correlated with output improvements.
Make a specific subscription or workflow decision
Write down the work you reviewed, the cost basis and the change you recommend:
- Renew a subscription when the engineer uses it for useful work at a cost the team can justify.
- Increase a usage limit when the current cap interrupts useful tasks. Record those tasks and review the result after the increase.
- Remove an unused paid seat after confirming coverage, actual use and the engineer’s needs.
- Change the workflow or provide training when repeated attempts, rework or review delays explain the problem.
Name the paid charge affected, if there is one, and set a follow-up date. Compare costs, output, quality and delivery again, keeping changes in assignments and responsibilities visible.
Use the AI Coding Productivity Review Worksheet to record the baseline, recommendation and follow-up. To see the process in Weave, book a demo and ask to walk through one engineer’s costs, completed PRs and code quality.
Sources
Measure AI coding productivity with Weave
See one engineer’s AI spending, completed code and quality in the same review. Open the PRs and cost details before deciding what to renew or change.
Book a demo