third-party agent

A third-party agent is an AI agent made by one company that runs and operates on a platform built by a different company. It's like a department store: the company that runs the store is different from the companies that operate the individual shops inside it. Here, the platform provider and the agent developer are two separate companies.

For example, agents from other companies like Claude or Codex can run inside a tool like GitHub Copilot. Copilot didn't build these agents itself, but it still needs some way to track and manage what they're doing on its platform.

The problem is this: if multiple shops (agents) are operating in the same space but you only see total sales—not which shop sold what—it's hard to decide which shop deserves more support. As teams increasingly use two or more agents at once, breaking activity down by agent has become a practical necessity.

The article describes a new feature that reports third-party agent app activity broken down by individual agent. One easy misunderstanding to avoid: this doesn't mean GitHub Copilot built agents like Claude or Codex itself. Copilot is simply the platform that runs them—each agent is developed by a different company.

If you have admin access, check the response from the Copilot usage metrics API. Look for a new array field that lists each agent's name, identifier, and execution count separately. If your organization uses multiple agents at once, this data lets you compare which agent is actually being used more.