How to Track AI Coding Token Spend by Team and Engineer with Weave
How to Track AI Coding Token Spend by Team and Engineer with Weave
Start by separating what you pay for AI coding tools from what those tools report about usage. Weave Token Intelligence brings together provider billing, usage telemetry, and configured subscription costs. It shows spend by engineer for supported sources and lets you compare teams while keeping the original cost labels visible.
Connect the tools that drive the bill
List every coding tool, subscription, and direct API account in scope. Start with whichever tools make up most of the bill. Weave's AI tool cost documentation lists the source for each integration. Claude Code, Cursor, GitHub Copilot, and Codex support per-engineer cost reporting where the relevant source data is available. Modal provides organization-level metered costs. Coverage depends on the account and integration.
Check the cost source before comparing teams
| Cost source | What to check |
|---|---|
| Provider or tool billing | Confirm the account, time period, and charges the integration actually reports. |
| Configured subscription | Check seat counts and who was assigned a seat during the period. |
| Modeled usage or client-reported cost | Read the source label and check whether the amount is an API-equivalent estimate or client-reported value. |
| Organization-level charges | Keep shared or unattributed charges visible alongside team totals. |
If a coding subscription includes usage, check whether those tokens are already covered by the seat fee before adding any usage estimate to your spend total.
Investigate a change in the bill
Pick one team and one billing period. Check that tool accounts map to the right engineers and teams. Compare the current period with the previous one by tool and, where available, by model. If Claude Code costs rose, check whether the source shows more usage, a change in model mix, additional seats, or overage charges. A change in coverage can also move the reported total.
For example, if the provider charged a hypothetical $1,200 and $200 of it has no person-level attribution, show $1,000 in attributed costs and $200 in shared charges. This keeps the team view honest and reconciles it to the full $1,200.
Read spend beside shipped work
Check output and review data for the same people, repositories, and dates. Weave uses estimated expert effort to describe the work represented by merged pull requests. Review cycles and reverts add quality context. The AI ROI methodology explains which costs and people are included in the public cost-per-output-hour measure and how the ratio is calculated.
Start with one team's last complete month. Reconcile the billed total, flag any gaps, and open a few of the pull requests behind the output trend before using the report for a renewal or budget decision.