Skip to main content

Decision brief

GitHub Copilot Review: What to Settle Before Standardizing on Copilot

Standardize on Copilot when GitHub is your system of record, after budgets, model policies and agent access are set. Paid usage past the pooled AI credits is on by default.

Reading time
3 min read
Sources
25 cited
In this piece
GitHub Copilot+15 Mixed

GILD verdict

Standardize on Copilot if your code and reviews already live on GitHub.com.

One vendor then covers IDE assistance, local and cloud agents, and pull request review under the organization policies and audit log you already run. Set budgets and agent policies before seats spread, because usage-based billing grows with adoption.

GitHub Copilot

AI Coding Agent by GitHub

+15 Mixed
Product page

Fit and limits

Who it fits

  • Your repositories and pull requests live on GitHub.com; Copilot Enterprise is designed for GitHub Enterprise Cloud.
  • Engineers should keep their current IDE. Copilot runs in VS Code, Visual Studio, JetBrains IDEs, Eclipse, and Xcode, and every plan includes the CLI and the Copilot desktop app.
  • You want to assign issues to a cloud agent that returns draft pull requests under your branch protections, and to run Copilot code review on every pull request.
  • You want models from OpenAI, Anthropic, Google, xAI, Microsoft, and Moonshot AI on one bill, with model policies per organization, and instructions in copilot-instructions.md or AGENTS.md.

Where it falls short

  • Copilot is not available for GitHub Enterprise Server, and the cloud agent works only on repositories hosted on GitHub.
  • Paid usage past the pooled AI credits is on by default for organizations, and spending limits only send alerts until "Stop usage when budget limit is reached" is enabled.
  • Each cloud agent session runs for at most 59 minutes, in one repository, on one branch, and opens one pull request.
  • Content exclusion does not apply to agent mode in IDE chat, and Copilot code review picks its model automatically without disclosing it.

Implementation reality

  1. Set budgets before you assign seats

    Business contributes 1,900 AI credits per user and Enterprise 3,900 to the billing entity's pool. Usage beyond the pool costs US$0.01 per credit unless an admin disables AI credits paid usage. GitHub recommends a universal user budget above the seat value and an enterprise spending limit. Enable "Stop usage when budget limit is reached" on every spending limit. Otherwise, limits only send alerts.

  2. Decide which models each organization may use

    AI credits are priced per model and token, so model policy is cost policy. GPT-6 Astra output costs US$50 per million tokens. GPT-6 Luna costs US$0.50 and Claude Opus 5.5 US$20 per million output tokens. Claude Fable 5 and 5.1 stay disabled on Business and Enterprise until an admin enables them, because GitHub's data retention agreement does not cover them. GitHub retired Claude Opus 4.7 on Oct 2, 2026 and Opus 4.5, Opus 4.6 and Sonnet 4.5 on Sep 1, 2026, so remove them from any policy or instructions that name them. Copilot code review selects its own model and also consumes GitHub Actions minutes. GitHub estimates a review at about US$0.05 to US$1 of AI credits on Lite effort and US$0.25 to US$5 on Balanced.

  3. Enable agents for a pilot first

    The cloud agent and third-party MCP servers are off by default for Business and Enterprise members. Enable them for one pilot organization first. Opt sensitive repositories out and retain the firewall and recommended allowlist. The agent pushes only to its copilot/ branch and cannot approve or merge its pull requests. A person with write access must approve workflows.

  4. Settle data and IP terms with legal

    GitHub excludes Business and Enterprise data from model training. It does not retain IDE chat or completion prompts and suggestions. Other interfaces retain them for 28 days. Engagement data stays for two years. Since April 3, 2026, Microsoft no longer requires the public code filter for IP indemnity on GitHub offerings; it stays available as an option.

What to verify before you buy

Agents

  • Computer use: not yet generally available. It works in Copilot CLI and the Copilot app on macOS and Windows and lets Copilot operate desktop apps under each surface's tool approval settings.

What to measure in your pilot

  • GitHub prices credits per model and token but publishes no cost per accepted pull request. Log credits and merges for a month by model and agent, and restrict the costliest models per merge.
  • GILD found only GitHub's own figure that 71% of Copilot code reviews surface actionable feedback. For one month, record which comments engineers fix before merge, and keep automatic review where most are fixed.
  • A July 2026 study's model puts the estimated merge probability of Copilot's agent pull requests at 59.6% in 309 popular Python repositories with 100+ stars, PRs to 20 November 2025, against 874 of 1,259 merged raw; over four weeks, count cloud agent pull requests on your largest repository that merge without follow-up commits.

Sources

Official documentation
21
Public source
4
  1. GitHub Copilot plans & pricingOfficial documentationgithub.com
  2. GitHub Copilot docs: PlansOfficial documentationdocs.github.com
  3. GitHub Copilot docs: Usage-based billing for organizations and enterprisesOfficial documentationdocs.github.com
  4. GitHub Copilot docs: Models and pricingOfficial documentationdocs.github.com
  5. GitHub Copilot docs: Getting started with budget controlsOfficial documentationdocs.github.com
  6. GitHub Copilot docs: Where to use GitHub CopilotOfficial documentationdocs.github.com
  7. GitHub Copilot docs: Copilot on GitHub.com and the cloud agentOfficial documentationdocs.github.com
  8. GitHub Copilot docs: Managing access to the cloud agentOfficial documentationdocs.github.com
  9. GitHub Copilot docs: Cloud agent risks and mitigationsOfficial documentationdocs.github.com
  10. GitHub Copilot docs: Copilot code reviewOfficial documentationdocs.github.com
  11. GitHub Copilot docs: Copilot CLIOfficial documentationdocs.github.com
  12. GitHub Copilot docs: GitHub Copilot appOfficial documentationdocs.github.com
  13. GitHub Copilot docs: Policies for enterprises and organizationsOfficial documentationdocs.github.com
  14. GitHub Copilot docs: Content exclusionOfficial documentationdocs.github.com
  15. GitHub Copilot docs: Repository custom instructionsOfficial documentationdocs.github.com
  16. GitHub Copilot docs: Reviewing audit logsOfficial documentationdocs.github.com
  17. GitHub docs: Copilot with data residencyOfficial documentationdocs.github.com
  18. GitHub docs: Terms for additional products (GitHub Copilot)Official documentationdocs.github.com
  19. GitHub Copilot docs: Supported AI modelsOfficial documentationdocs.github.com
  20. Microsoft Learn: Customer Copyright Commitment required mitigations (GitHub change effective Apr 3, 2026)Official documentationlearn.microsoft.com
  21. GitHub Blog: 60 million Copilot code reviews and counting (Mar 5, 2026)Public sourcegithub.blog
  22. arXiv: Adoption and Impact of Command-Line AI Coding Agents, Microsoft's early 2026 rollout (Jul 1, 2026)Public sourcearxiv.org
  23. arXiv: How Do AI Coding Agents Contribute to Software Development? An Empirical Study of Agentic Pull Requests (Jul 23, 2026)Public sourcearxiv.org
  24. arXiv: The Rise of AI Teammates in Software Engineering 3.0, the AIDev dataset (Jul 2025)Public sourcearxiv.org
  25. GitHub Copilot docs: Computer useOfficial documentationdocs.github.com

Keep reading

  1. Buyer guideMoving Off Continue: A Migration ChecklistInventory a team's Continue setup and carry its models, rules, MCP servers, and prompts to a maintained tool, and run a pilot before removing Continue.
  2. Buyer guideGitHub Copilot Workflow PlaybookRepository instructions, agent-ready issues, and a review routine for cloud agent pull requests.
  3. Buyer guideGitHub Copilot AI Credit Budget & Policy ChecklistThe budget controls to set under AI credits, in GitHub's order, sized with GitHub's method, plus model and code review billing policy.
Read the research