GitHub Copilot Agent Metrics Bug Undercounts IDE Usage: Update VS Code 1.139.0

The fix arrives through IDE and plugin updates. Visual Studio Code already has it. Visual Studio, JetBrains, Eclipse and Xcode get it over the next several weeks. GitHub is clear that agent activity your reports already missed will not come back.

What broke, and who it affects​

GitHub says the problem only affects IDE versions that use the Copilot SDK for agent mode. Developers on older versions are still counted normally. The affected sessions came from inside an IDE but carried no IDE identity, so the reporting system had no editor to assign them to.

The fields hit hardest are the agent lines-of-code totals. GitHub names loc_added_sum and loc_deleted_sum under the agent_edit feature value. GitHub's metrics reference describes agent_edit as the lines Copilot writes directly into files in agent and edit mode, including edits from custom agents. That makes it different from suggested lines and from accepted completions. According to the changelog, agent interactions and agent lines of code stay undercounted in:

• Enterprise reports
• Organization reports
• User reports
• Both 1-day and 28-day reporting windows

What it doesn't affect: your bill. Billing is unaffected, missing data cannot be backfilled, and some Copilot CLI metrics may be inflated until clients update.

Summary: The SDK change cut the link between agent sessions and the IDE they ran in. Agent metrics dropped as a result, and some of that activity was counted as CLI use instead.

The fix schedule, IDE by IDE​

GitHub published this availability table on October 6. Only the VS Code release is available now. The rest are expected dates.

Visual Studio and VS Code are Microsoft's own editors, so Windows-heavy shops get the fix first. VS Code users can update today. Visual Studio users need to watch for 18.12. GitHub says it will add each fixed version to its supported-IDE documentation as it ships, and expects the whole rollout to finish by November 2026.

Summary: Update to VS Code 1.139.0 or later now. Then watch for Visual Studio 18.12 and the next JetBrains, Eclipse and Xcode plugin releases.

Missing data is gone for good​

Lost agent data stays lost, and that should shape how you plan. GitHub says it can't backfill the gap because activity from affected versions never said which IDE it came from, so it can't be reassigned later. Developers must update because missing data from older versions cannot be backfilled afterward.

That has three practical effects:

1. Every day a developer stays on an affected version is a day of agent data you won't get back. Updating stops the losses. It doesn't recover what's already missing.
2. Recovery will be gradual. GitHub expects agent numbers to climb back as developers update, not jump in one go. A slow rise over the next few weeks tells you how far your rollout has got. It doesn't mean developers suddenly started using agents more.
3. Copilot CLI numbers from this period are likely inflated. GitHub says CLI figures will settle as SDK-based clients update. Earlier CLI figures can't be corrected, because the misattributed activity can't be separated from real CLI use. People who only use Copilot CLI don't need to update anything for this fix.

If you've shown leadership a chart of "CLI adoption up, agent mode down," add a note to it. Some of that movement was really one surface being counted as another.

What admins should do now​

GitHub's advice comes down to managing your developers' environments. Here's a practical order:

1. Find out who's on which version. In per-user reports, each totals_by_ide entry includes last_known_ide_version and last_known_plugin_version. GitHub's docs describe these as the most recently detected versions. They're good for checking coverage, but they aren't a live inventory of every machine.
2. Push VS Code 1.139.0 or later now. If you control versions centrally, GitHub suggests moving developers straight to the fixed releases so they skip any intermediate builds.
3. Queue the other IDEs. Get approval and packaging ready for Visual Studio 18.12 and the next JetBrains, Eclipse and Xcode plugins, so you can deploy as soon as they ship.
4. Set minimum versions in your device management tooling wherever you can.
5. Check telemetry settings and network paths. IDE telemetry must stay on, and proxies and firewalls must allow the Copilot telemetry endpoint.
6. Mark the change in your trend reports. Record when the fixed versions went out so anyone reading quarter-over-quarter numbers knows about the gap.

One more caution: GitHub's reconciliation docs say IDE telemetry is processed asynchronously and usually finalizes within three full UTC days. Don't judge the update's effect on yesterday's numbers.

Summary: Check versions, update VS Code now, line up the other fixed releases, and annotate your reports.

Why client telemetry keeps causing trouble​

This bug points to a larger weakness. GitHub explains that the detailed breakdowns in usage metrics, including feature, language, model and lines of code, come from telemetry each IDE sends. Server-side records reliably show who was active, but they can't see what happened inside the editor. GitHub lists the usual causes of missing client data:

• Telemetry is turned off in the IDE
• A proxy or firewall blocks the Copilot telemetry endpoint
• The IDE or Copilot extension is out of date and doesn't send the events metrics rely on
• The client changed how it sends telemetry, as happened with the SDK move
• The editor is unsupported or third-party and doesn't send Copilot telemetry

GitHub has been using more server-side data to cover these gaps. WindowsForum covered a July 2, 2026 API update in which users observed only through server-side telemetry now have IDE and plugin versions surfaced in the totals_by_ide breakdown. GitHub now says usage metrics use server-side data to count active users that client telemetry misses and to identify their IDE. It also concedes that lines of code and accepted suggestions can only come from the editor, so client-side gaps won't go away entirely. The result: a user can appear in your active-user totals with empty feature and lines-of-code breakdowns.

Don't confuse this bug with the dedicated VS Code Agents window metrics. GitHub added those in September, saying reports now include generally available metrics for activity in the dedicated VS Code Agents window. GitHub's reference docs say those fields are separate from editor-window agent mode, including used_agent and agent_edit, and aren't included in the general totals. So Agents-window numbers won't show this gap, and they won't fill it either.

The bigger picture​

This is the latest in a run of fixes to Copilot's usage metrics. Over the past year GitHub has adjusted how chat is counted, added CLI lines-of-code data and fixed AI-credit attribution. Each fix makes the numbers more accurate. Each also reminds admins that these dashboards are estimates built from telemetry, and the clients sending that telemetry keep changing.

To GitHub's credit, it explained this problem plainly. It named the cause, listed the affected fields, ruled out any billing impact and said up front that the lost data can't be recovered. It's still awkward that moving agent sessions to the SDK took IDE identification with it. Organizations using these metrics to justify AI spending have been looking at an understated agent picture without knowing it.

If Copilot data feeds your budget decisions, make IDE version management part of keeping those numbers accurate. Agent use probably never dropped. Your reports just weren't recording all of it.