BurntAI

Per-developer AI spend tracker: how to build one from GitHub and Anthropic data

By Jiri Sifalda

A per-developer AI spend tracker answers one question at the monthly review: what did each person cost across every AI tool, and is that inside their budget? The provider consoles do not answer it. Each console shows its own product, in its own unit, for its own set of users.

This guide shows where the per-person data lives for GitHub Copilot and Claude Code, what each source leaves out, and how to join them into one row per developer.

Why the seat price is not the per-developer cost

GitHub lists Copilot Business at $19 per granted seat per month and Copilot Enterprise at $39 per granted seat per month (GitHub Copilot plans). That is the floor, not the bill.

Under usage-based billing, each plan includes GitHub AI Credits: 1,900 per user per month on Business and 3,900 on Enterprise, at 1 AI credit = $0.01 USD (usage-based billing for organizations and enterprises). Three details on that page change how you track spend per person:

  • Credits are pooled across the organization or enterprise. A heavy user draws on credits that a light user did not spend, so the invoice does not show who used what.
  • Additional usage is enabled by default on Business and Enterprise. When the pool runs out, usage continues at published per-credit rates and is charged to the organization.
  • Inline code completions and next edit suggestions are not billed in AI credits. Two developers with the same completion volume can have very different credit use, because chat and agent work is what consumes credits.

So the per-developer number you need is credits consumed, not seats granted.

Where the per-person Copilot data lives

GitHub's Copilot usage metrics APIs provide enterprise- and organization-scoped reports with user-level detail (About Copilot usage metrics). Since June 19, 2026, the user-level reports (users-1-day and users-28-day) include an ai_credits_used field for each user (GitHub changelog).

Read that field with care:

  • It is an overall per-user total. It is not broken down by feature, model or surface.
  • GitHub calls it "a metrics signal for analyzing consumption, not a billed total". Use it to attribute spend, and reconcile the total against your invoice.
  • Only enterprise administrators and organization owners with access to Copilot usage metrics can read it through the REST API.
  • Data for a given day is available within two full UTC days after that day closes (About Copilot usage metrics).

The endpoints themselves are documented in the Copilot usage metrics REST reference.

Where the per-person Claude Code data lives

For a Claude Console organization, the Claude Code Analytics Admin API returns daily metrics per actor from /v1/organizations/usage_report/claude_code (Claude Code Analytics API). Each record names the actor by email address, or by API key name for key-based use, and carries a model breakdown with token counts and an estimated_cost.amount in cents USD.

The same page lists the limits that matter for a tracker:

  • It needs an Admin API key, an OAuth token with the org:admin scope, or an unscoped key. Workspace API keys do not work.
  • It returns one UTC day per request, and data appears with up to a 1-hour delay.
  • An actor can have several records for one day, one per terminal type and account type. Add them up to get the daily total.
  • It is free to use for organizations with Admin API access.
  • It does not return usage from Claude subscription plans such as Pro, Max, Team or Enterprise. That usage belongs to the claude.ai organization. Anthropic points to the Claude Enterprise Analytics API, the claude.ai analytics dashboard, or the OpenTelemetry export for those users.

If some developers sign in with a subscription and others use your Console organization, one API will not show everyone. List which developers use which path before you trust a total.

How to join the two sources into one row per developer

  1. Pick one identity key. GitHub reports a login and Anthropic reports an email address. Keep a mapping table from GitHub login to work email, and own it in one place.
  2. Pull both sources daily. Store GitHub's ai_credits_used per user per day and Anthropic's estimated_cost.amount per actor per day, summed across records.
  3. Convert to one currency unit. Multiply credits by $0.01. Divide Anthropic cents by 100.
  4. Add the seat floor. Add each person's seat price, so a developer who used no credits still shows a cost.
  5. Respect the delays. GitHub data can lag by two UTC days and Anthropic data by an hour. Mark the last complete day in the report, so nobody reads a gap as low usage.
  6. Compare to a budget. Put each person's month-to-date total next to their budget and the share of the month that has passed.
  7. Reconcile monthly. Both per-user numbers are estimates. Compare the summed totals against each invoice and record the difference.

The result is the table an engineering leader actually wants: one row per developer, Copilot and Claude in separate columns, a total and the remaining budget.

What a tracker cannot tell you

A tracker shows who spent what. It does not show whether the spend was worth it. Claude Code's analytics add sessions, commits and pull requests per actor, and GitHub's reports add activity per user, so you can put output next to cost. Treat those as context for a conversation with the developer, not as a score.

Skip the plumbing

BurntAI is built to produce this table for you: total AI spend in one place, per developer, across providers, against a budget. See pricing to get on the list, or read more on the blog.

See every dollar your team spends on AI

See every developer's GitHub Copilot and Claude spend in one dashboard. Track per-person AI costs, set budgets, and catch overspend before the invoice lands.