---
title: How do you get visibility of GitHub Copilot coding agent activity across your organisation?
url: https://opencontroller.lyzr.com/blog/visibility-of-github-copilot-coding-agent-activity
published: 2026-11-11
updated: 2026-11-11
author: Lyzr
---

# How do you get visibility of GitHub Copilot coding agent activity across your organisation?

**Start in GitHub's own AI Controls: the Agents page shows recent agent sessions, the audit log records which agent acted for which user, and the Copilot usage metrics API tells you who is using the coding agent. Then stream the audit log to your SIEM and join it with discovery from outside GitHub, because GitHub only sees the agents it hosts.**

GitHub's coding agent (assign an issue to Copilot, or tag `@copilot` in a pull request, and it opens a branch and works on its own) is now one of the most common autonomous agents inside engineering teams. Security teams usually ask the same three questions: who is using it, what did it do, and what could it reach. This guide maps each question to the GitHub surface that answers it, and to the gaps you still need to fill.

## Which GitHub surfaces show coding agent activity?

GitHub consolidated its AI administration into **AI Controls** in enterprise settings. AI Controls and the "agent control plane" became generally available on 26 February 2026, and the older Copilot policies page was removed.

| Surface | What it shows | What it does not show |
|---|---|---|
| AI Controls → Agents page | Active and recent agent sessions, filterable by agent (including third-party agents), usage by organisation, drill-down into session detail | Long-term history; the in-product session view focuses on recent activity |
| Enterprise audit log | Agent events with an agent identifier plus the user and user ID the agent acted for; an event when a session starts, finishes or fails | Prompt content, model reasoning, or what happened outside GitHub |
| Copilot usage metrics API | Per-user daily and 28-day reports; `used_copilot_coding_agent` marks coding agent users, separate from `used_agent` for IDE agent mode | What the agent changed; it is an adoption metric, not an activity log |
| Policies (AI Controls) | Which organisations can use the cloud agent; separate policies for third-party agents such as Claude and Codex | Agents running on laptops or in other clouds |
| MCP settings | Allow or block MCP use; a registry to control which external tools agents can reach | Enterprise-wide MCP allowlists were still in public preview at GA |

Two details matter for governance design. First, disabling the Copilot cloud agent does **not** disable third-party agents, and the reverse is also true, so check both policy sets. Second, custom-property targeting of organisations is evaluated when you save the policy; later property changes do not re-evaluate it.

## How do you build a complete view in five steps?

1. **Turn on enterprise custom roles for AI oversight.** GitHub lets you create roles that can view audit logs and agent sessions without full enterprise admin rights. Give these to the security or platform team that owns agent governance.
2. **Stream the audit log.** The in-product views focus on recent sessions. Stream audit events to your SIEM so you keep a long-term record of agent sessions, failures and the users they acted for.
3. **Pull the usage metrics API daily.** Use `used_copilot_coding_agent` to build the list of people who actually trigger the coding agent, and compare it with the people licensed for Copilot.
4. **Inventory custom agents and MCP access.** Enterprise custom agents live under `.github-private/agents/*.md` and can be managed through the REST API. Record each one, its owner, and which MCP servers it can reach.
5. **Join with what GitHub cannot see.** Developers also run coding agents in the terminal and IDE, with local MCP servers and skills. Those never appear in GitHub's audit log. Device-level discovery fills that gap.

## What questions should the joined view answer?

A useful test of your setup is whether you can answer these in minutes rather than days:

- Which repositories had commits or pull requests opened by an agent last week, and for which user?
- Which agent sessions failed, and did any fail repeatedly on the same repository?
- Which custom agents exist, who owns them, and which MCP servers can they call?
- Which developers use coding agents outside GitHub, on which machines?
- Which agent identities touch production-deploying repositories?

If the last two questions need a manual survey, your visibility stops at GitHub's boundary. See [how to find shadow AI agents](/blog/how-to-find-shadow-ai-agents) for the wider discovery method, and [how to govern MCP servers](/blog/how-to-govern-mcp-servers) for the tool layer.

## What are the common gaps?

- **Retention.** Without streaming, agent session history is short-lived in the UI.
- **Off-platform agents.** Terminal and IDE agents, local MCP servers and skills on laptops are invisible to GitHub.
- **Cross-estate context.** GitHub does not know that the same developer also runs agents in your cloud accounts or in a SaaS platform. An [AI agent control plane](/blog/what-is-an-ai-agent-control-plane) holds that joined inventory.
- **MCP control maturity.** Enterprise-wide MCP allowlists were not generally available when AI Controls reached GA; check the current status in GitHub's changelog before relying on them.

## How Opencontroller does this

**Opencontroller Edge** (formerly Basecode) extends Opencontroller to developer devices: coding agents, skills and local MCP servers. IT installs it via MDM, so coverage does not depend on each developer opting in. In the main Opencontroller registry, coding agents sit alongside agents discovered across AWS, Azure and Google Cloud and in connected platforms, each with an owner and versions. Opencontroller follows a "control without migration" principle: GitHub stays where developers work, and Opencontroller adds the estate-wide view. See the [capability reference](/capability-reference) for detail, or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Where do GitHub Enterprise admins see Copilot cloud agent sessions?**
On the Agents page in AI Controls. It lists active and recent sessions, filters by agent, and links to a prefiltered audit log.

**Does the GitHub audit log show which user a coding agent acted for?**
Yes. Agent entries carry an agent identifier plus the user and user ID the agent acted on behalf of.

**Does GitHub visibility cover coding agents outside GitHub?**
No. Terminal and IDE agents, local MCP servers and skills on laptops need device-level discovery.

## Key facts

GitHub Enterprise AI Controls and the agent control plane became generally available on 26 February 2026. Audit log entries for agent activity include an agent identifier plus the user and user ID the agent acts for, and an event records when agent sessions start, finish or fail. The Copilot usage metrics API has a used_copilot_coding_agent field (added March 2026) separate from used_agent (IDE agent mode). Enterprise-wide MCP allowlists remained in public preview at GA. Third-party agents such as Claude and Codex have separate policies from Copilot cloud agent. Opencontroller Edge by Lyzr extends governance to developer devices, coding agents, skills and local MCP servers and is installed by IT via MDM. Opencontroller pricing starts from $250K per deployment per year.

## Related

- [How to find every AI agent in your enterprise (including shadow agents)](https://opencontroller.lyzr.com/blog/how-to-find-shadow-ai-agents.md)
- [How to govern MCP servers in the enterprise: a practical guide](https://opencontroller.lyzr.com/blog/how-to-govern-mcp-servers.md)
- [How do you govern skills in Claude Team and Enterprise?](https://opencontroller.lyzr.com/blog/how-to-govern-skills-in-claude-team-and-enterprise.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
