Windows 11 Is Counting Which Coding Agents You Run
Microsoft sells Copilot, and according to one researcher's teardown, Windows 11 also keeps a tally of whether you run Claude Code, Codex, Cursor or Antigravity. The post went up on xusheng.dev dated Nov 16 and hit Hacker News late that day, reaching 702 points and 580 comments in this window, the biggest agent-adjacent thread of the day. Post: https://xusheng.dev/posts/windows-coding-agent-telemetry/
What the researcher found, by reverse engineering MissionControlAggregator.dll: a feature called CLICodingAgentUsage with executable-name matchers. Four slots for CLI agents (copilot.exe, claude.exe, codex.exe, agy.exe, plus their extensionless WSL versions), five for desktop agents including GitHub, Claude, Codex, Cursor and Antigravity, and more for developer tools like python, dotnet, gcc and VS Code. Usage gets stored as bits, one per tool, and emitted as a SubCohorts event. The researcher then found that event sitting in the Windows diagnostics upload queue when optional diagnostic data is on, and on a clean US Windows 11 Home install that toggle defaults to on. The test was simple: rename cmd.exe to gcc.exe and watch a slot flip. The researcher also says an AI agent helped find it.
Two things keep this from being a scandal headline. What leaves the machine, per the analysis, is which tools were used, not code or prompts. And it stops if you turn optional diagnostic data off. No Microsoft statement had appeared by the end of the window. Still, the company with its own coding agent in the editor now has a population-level view of how often its rivals get launched, on Ignite day of all days.
Apple moved the other way the same night. The macOS 27.3 beta now warns against giving AI apps full access and nudges users toward choosing specific apps instead. Apple note: https://support.apple.com/en-us/149699
So the operating systems have noticed the agents. One is counting them, the other is fencing them. If you ship a coding agent, the OS is no longer a neutral floor, and your binary name is a data point in someone else's market share chart.
← Back to all articles
What the researcher found, by reverse engineering MissionControlAggregator.dll: a feature called CLICodingAgentUsage with executable-name matchers. Four slots for CLI agents (copilot.exe, claude.exe, codex.exe, agy.exe, plus their extensionless WSL versions), five for desktop agents including GitHub, Claude, Codex, Cursor and Antigravity, and more for developer tools like python, dotnet, gcc and VS Code. Usage gets stored as bits, one per tool, and emitted as a SubCohorts event. The researcher then found that event sitting in the Windows diagnostics upload queue when optional diagnostic data is on, and on a clean US Windows 11 Home install that toggle defaults to on. The test was simple: rename cmd.exe to gcc.exe and watch a slot flip. The researcher also says an AI agent helped find it.
Two things keep this from being a scandal headline. What leaves the machine, per the analysis, is which tools were used, not code or prompts. And it stops if you turn optional diagnostic data off. No Microsoft statement had appeared by the end of the window. Still, the company with its own coding agent in the editor now has a population-level view of how often its rivals get launched, on Ignite day of all days.
Apple moved the other way the same night. The macOS 27.3 beta now warns against giving AI apps full access and nudges users toward choosing specific apps instead. Apple note: https://support.apple.com/en-us/149699
So the operating systems have noticed the agents. One is counting them, the other is fencing them. If you ship a coding agent, the OS is no longer a neutral floor, and your binary name is a data point in someone else's market share chart.
Comments