The AI week, distilled.
Week 38 · 2026
Microsoft AI | 14–20 September 2026

GitHub Copilot adds spending controls and agent reporting ahead of October model retirements.

GitHub introduced cost-and-quality tiers for Copilot and budget increase requests for eligible enterprise customers. It also expanded agent usage reporting and announced model retirements for 19 October, giving development leaders a month to review model policies and test replacements.

01

GitHub schedules Copilot model retirements for 19 October

GitHub announced that selected Copilot models will retire on 19 October across chat, editing, agent modes and code completions. It provided suggested replacement models.

  • Inventory model-dependent workflows and test replacement models before 19 October.
  • Review Business and Enterprise model policies because GitHub automatically enables replacements under default settings unless administrators override them.
  • Benchmark replacement models on internal coding tasks before updating developer guidance and spending forecasts.
02

Copilot auto selection adds cost-and-quality tiers

GitHub introduced efficiency, balance and intelligence tiers for Copilot auto model selection. The rollout covers VS Code, Copilot CLI and the GitHub Copilot app.

  • Define when developers should prioritize cost, response time or output quality rather than applying one setting to every task.
  • Track actual model charges during pilots because GitHub bills auto usage according to the model selected.
  • Review the enabled model set centrally because the three tiers select from the same underlying pool.
03

Copilot budget increase requests reach general availability

Copilot Business and Enterprise users on usage-based billing can request additional budget after exhausting their AI credits. Organization owners, enterprise owners and billing managers can approve, adjust or deny requests in settings.

  • Assign approvers and response targets before rollout so exhausted credits do not leave developers waiting for an undefined escalation process.
  • Use adjustable approvals to allocate additional spending to priority projects rather than raising every user's budget.
  • Check tenant eligibility because GitHub excludes enterprises with managed users from this feature.
04

Copilot CLI metrics expand to agent customizations

GitHub expanded its usage metrics API to cover Copilot CLI agent customizations. Reporting now includes skills, custom agents, Model Context Protocol servers, slash commands and plugins.

  • Add customization usage to engineering dashboards to distinguish basic CLI adoption from use of enterprise-specific agents.
  • Compare reported MCP server and plugin usage with the tools your organization has approved.
  • Use adoption data to target training and review unused customizations rather than treating usage counts as proof of productivity.
05

Copilot weekly update adds Sentry and Dev Container integrations

GitHub's weekly release update highlighted Sentry integration in the Copilot app and deeper agent integration with local Dev Containers. It also reported generally available VS Code agent usage metrics and code review enhancements.

  • Review what incident data agents can access before connecting Sentry to enterprise development workflows.
  • Pilot Dev Container integration against approved development images to check environment compatibility before expanding agent access.
  • Use VS Code agent session and message metrics for adoption reporting, with separate measures for code quality and delivery outcomes.

Copilot changes this week

From Message Center and the public Microsoft 365 roadmap — admin actions first.

All 56 Copilot changes on ai-news.cz →

Microsoft Learn: what changed

The most significant documentation changes this week.

All 359 documentation changes →