Our verdict · Coding agent
The cheapest way to put a coding agent on a team, and the most natural if your work already lives in GitHub issues and pull requests: assign an issue to Copilot and it opens a PR for review. The limits are real: one repository and one PR per task, a hard 59-minute session cap, and GitHub-hosted repos only.
Pick it if
- Teams whose work starts as GitHub issues
- Organisations that want admin-controlled, pooled usage
- Well-scoped tasks: tests, small features, fixes, docs
Skip it if
- Your code isn't on GitHub
- Tasks span several repositories or run longer than an hour
- You want to build a new app from a description
GitHub Copilot started as autocomplete and is now a family of agents. This review is about the one that changes how a team works: Copilot cloud agent (formerly the coding agent), which takes an issue and returns a pull request. You assign an issue to Copilot, it works in its own environment, and it opens a PR with you as reviewer; mention @copilot in the PR to ask for changes. This review compares GitHub's published plans and documentation against our six criteria.
Scores
- Speed to a first version2/5
Not a builder: it works through issues and PRs on an existing repository.
- Depth before the wall3/5
Runs tests in a real environment, but one repo per task and a 59-minute cap limit larger changes.
- Backend and data2/5
Brings no backend; it works on your repo's stack.
- Ownership and exit5/5
Output is commits and PRs in your own repository.
- Pricing predictability4/5
Published seat prices, published credit allowances and a published $0.01 overage rate for organisations; Actions minutes on top.
- Team and production readiness5/5
Org-wide pooled credits, admin controls, rulesets and branch protections, native PR review.
Overall 3.5/5, the plain average of six criteria. Scored by Mahmoud Halat from published specs and docs, checked Nov 1, 2026. How we score.
Pricing: AI credits and Actions minutes
| Plan | Price (USD) | What you get |
|---|---|---|
| Free | $0 | Limited chat and agent usage; no cloud agent |
| Pro | $10/mo | 1,500 AI credits (1,000 base + 500 flex), cloud agent |
| Pro+ | $39/mo | 7,000 AI credits (3,900 base + 3,100 flex) |
| Max | $100/mo | 20,000 AI credits (10,000 base + 10,000 flex) |
| Business | $19/seat/mo | 1,900 AI credits per user, pooled; admin must enable the agent |
| Enterprise | $39/seat/mo | 3,900 AI credits per user, pooled |
- Billed in
- AI credits (1 credit = $0.01) plus GitHub Actions minutes for agent sessions
- Free tier
- Limited chat and agent usage; no cloud agent
Source: GitHub docs: Copilot plans, Plans page, About the cloud agent. Prices checked Nov 1, 2026, before tax. On Business and Enterprise, usage beyond the pool is $0.01 per credit where an admin allows it; individual plans can buy extra usage at the rates in GitHub's billing docs. The cloud agent also uses GitHub Actions minutes.
Since June 2026, Copilot bills in AI credits, each worth $0.01, instead of counting "premium requests". Individual plans have a base allowance plus a "flex" allotment of extra usage; Business and Enterprise pool every seat's credits across the organisation, with admins deciding whether usage beyond the pool is allowed at $0.01 a credit. GitHub gave existing Business and Enterprise customers higher allowances for the first three months of the new billing (3,000 and 7,000 credits a user until September 1, 2026).
The cloud agent also uses GitHub Actions minutes, because it runs in an Actions-powered environment. On a busy repository that's the second line on the bill, and it's easy to miss.
Even so, this is the most predictable pricing in our group: published seat prices, published credit allowances and a published overage rate. For a 20-person team on Business, the list price is $380 a month before Actions minutes.
How the agent works
Copilot cloud agent runs in "its own ephemeral development environment, powered by GitHub Actions", where it can explore the code, make changes and run tests and linters. You can start it from an issue's sidebar, from the repository's Agents tab (which also takes non-coding research tasks), from VS Code, or from Slack and Microsoft Teams. Jira, Linear and Azure Boards can create PRs through it too, though deeper planning and iteration stay on GitHub.
Because the output is an ordinary pull request, it goes through your existing reviews, checks and branch protections. That's the main reason teams adopt it: nothing about the review process changes.
Limits that shape the work
GitHub documents the limits plainly:
- It can only change the repository named when the task starts, and reads context only from that repository unless you configure MCP access.
- It works on one branch at a time and opens exactly one pull request per task.
- Each session has a hard 59-minute maximum that can't be extended.
- It only works with repositories hosted on GitHub, and rulesets that restrict commit authors can block it unless Copilot is added as a bypass actor.
Those limits make it excellent for well-scoped tickets (a failing test, a small feature, a dependency update, documentation) and a poor fit for cross-repository changes or long refactors. For those, an agent that runs in your own environment, such as Claude Code, goes further.
Ownership
The agent's work arrives as commits and a pull request in your own repository. There's nothing to export and nothing that lives inside Copilot.
Key facts
- Made by
- GitHub (Microsoft)
- Works in
- GitHub issues and PRs, VS Code, the Copilot CLI, Slack and Teams; Jira, Linear and Azure Boards can create PRs
- Runs in
- An ephemeral GitHub Actions environment that can run tests and linters
- Limits
- One repo and one PR per task, one branch at a time, 59-minute session cap
- Ownership
- Your GitHub repository
Alternatives
- Claude CodeLonger, multi-repo sessions and more surfaces
- CursorAn editor-first agent with background agents and Bugbot
- Windsurf (Devin Desktop)Agentic editor with cloud agents included on Pro
Questions people ask
Which Copilot plans include the coding agent?
All paid plans. GitHub's docs say Copilot cloud agent is available for all paid Copilot plans, from Pro at $10 a month. On Copilot Business and Enterprise an administrator has to enable it first. The Free plan doesn't include it.
How is the Copilot coding agent billed?
It uses AI credits and GitHub Actions minutes. Since June 2026 Copilot bills in AI credits worth $0.01 each; Pro includes 1,500 a month, Business 1,900 per user pooled across the organisation, and Enterprise 3,900. On Business and Enterprise, usage beyond the pool is $0.01 per credit where an admin allows it; individual plans can buy extra usage at the rates in GitHub's billing docs. Checked November 1, 2026.
What are the Copilot coding agent's limits?
GitHub documents several. It can only change the repository named when the task starts, works on one branch at a time, opens exactly one pull request per task, and has a hard 59-minute session limit. It only works with repositories hosted on GitHub.
Copilot or Claude Code for a team?
Copilot is cheaper per seat and lives where GitHub teams already work, which makes it a good default for well-scoped tickets. Claude Code handles longer, multi-repository work and runs in more places. Many teams use both.