What engineering AI leaves unanswered
Copilot seats are easy to buy and hard to account for. The recurring questions are:
- How many assigned Copilot seats are actually used each week?
- Which teams adopted it, and which have dormant seats we keep paying for?
- Is developer engagement growing, flat, or fading after rollout?
- What does Copilot cost per active developer?
- How does developer AI compare with assistant AI in the same budget?
Dormant developer seats are among the most common sources of avoidable AI spend, precisely because nobody owns the reconciliation between GitHub's admin view and the contract.
What GitHub provides — and where the boundary sits
GitHub exposes solid organization-level Copilot reporting: seat assignment state and daily metrics covering active and engaged developers. For understanding Copilot inside GitHub, that is the authoritative source, and the Midgentic connector reads exactly those endpoints.
What it cannot do is leave GitHub. It does not hold your negotiated per-seat price, it does not know that the same engineers also consume a model API, and it has no view of the assistant licenses the rest of the business runs. Reconciling those is a portfolio problem.
What Midgentic actually ingests from GitHub
- Copilot seat data
- Seat assignments for your GitHub organization, which is the denominator for utilization.
- Daily Copilot metrics
- Active and engaged developer counts as published by GitHub, retained as a historical series.
- Usage trends
- Adoption movement over time rather than a single point-in-time snapshot.
- Developer identity resolution
- GitHub users matched to people in your workspace so adoption can be read by team.
- License terms (entered by you)
- Copilot Business or Enterprise seats, price, currency, and effective dates — GitHub's metrics endpoints do not expose billing.
- CSV import
- Supplementary usage or cost exports where you want to bring in data the API does not provide.
No repository content, source code, pull request text, or suggestion content is collected.
The questions this data answers
- Developer adoption
- How many engineers use Copilot, and how consistently.
- Seat utilization
- Assigned seats against active developers, so dormant seats surface before renewal rather than after.
- Team-level patterns
- Which engineering groups genuinely adopted the tool and which need enablement.
- Cost per active developer
- License terms read against real activity, in your billing currency.
- Portfolio comparison
- Developer AI value set next to assistant and API investment instead of judged in isolation.
Engineering AI as part of one portfolio
Midgentic does not replace GitHub's Copilot dashboard and does not try to measure code quality or developer performance. It answers the funding question: what did we buy, who uses it, what is it costing per active developer, and how does that compare with everything else in the AI budget. That work continues in adoption analytics and spend intelligence.
Common questions
What GitHub Copilot data does Midgentic collect?
Copilot seat assignments for your GitHub organization and the daily Copilot metrics GitHub publishes: active and engaged developer counts and usage trends. Users are resolved so engineering adoption can be attributed to teams.
Does Midgentic read our source code or Copilot suggestions?
No. The connector uses GitHub's organization-level Copilot metrics and seat endpoints. It does not access repositories, source code, pull request contents, or the text of Copilot suggestions.
Does GitHub Copilot spend sync automatically?
No. GitHub's Copilot metrics endpoints report activity and seats, not billing. Copilot Business or Enterprise cost comes from license terms you enter, which Midgentic reads against seat and activity data to produce cost per active developer.
How is this different from GitHub's own Copilot dashboard?
GitHub's dashboard is the authoritative view of Copilot inside GitHub, and Midgentic reads from the same APIs. Midgentic's role is to place developer AI next to assistant AI, API platforms, and manually tracked tools, so engineering AI is part of one portfolio picture instead of a separate report.
Is the GitHub Copilot connector generally available?
It is implemented and in beta, enabled per workspace during onboarding while it is validated with early customers.