Note: The enterprise/platform research file was not available for this edition; coverage of enterprise features is drawn from the remaining research sources.

The Biggest News: Usage-Based Billing Is Here

GitHub Copilot officially moved to token-based billing on June 1, replacing Premium Request Units with GitHub AI Credits across all plans. Under the new model, every Copilot interaction consumes credits calculated on token consumption (input + output + cached) per model. Code completions and Next Edit Suggestions remain free โ€” everything else now costs: Copilot Chat, Copilot CLI, cloud agent sessions, code review, Copilot Spaces, and third-party agents.

The pricing structure lands as follows: Pro ($10/mo) gets 1,000 credits, Pro+ ($39/mo) gets 3,900, Business ($19/user/mo) gets 1,900, and Enterprise ($39/user/mo) gets 3,900 per user. One credit equals $0.01. A new Copilot Max tier offers higher credit allotments and spending limits for heavy agentic users, available as an upgrade to existing Student, Pro, and Pro+ subscribers.

The community reaction has been intense. Reports surfaced of single agentic prompts consuming 576 of 1,500 monthly credits on Pro plans, with multiple outlets โ€” TechCrunch, Tech Times โ€” covering developers reporting 10ร— to 50ร— cost increases for heavy agentic workflows. GitHub responded by releasing user-level budget controls (now GA), enabling admins to set universal or per-user budgets with email notifications as limits approach.

Why This Matters

This is the most consequential billing change in Copilot’s history. For enterprise teams, the switch from all-you-can-eat to consumption-based pricing fundamentally changes the economics of agentic development. Teams running aggressive agent workflows will need to budget carefully, choose cost-effective models, and monitor credit burn rates. On the positive side, the new user-level budget controls and cost-per-model display in VS Code give admins the tools they need to manage spend โ€” but the sticker shock for early adopters is real. If you’re planning an enterprise rollout, audit your team’s current agent usage patterns before the next billing cycle.

Claude Opus 4.8 GA โ€” Smarter Code Understanding

Anthropic’s Claude Opus 4.8 is now generally available within GitHub Copilot across all modes (chat, agent, inline) in every supported IDE. GitHub describes it as “a clear step forward in code understanding, particularly for navigating complex, large codebases.” It launched with a 15ร— premium multiplier that was in effect only until the June 1 billing transition.

For enterprise teams, this means the best available model for tackling sprawling monorepos and deeply nested dependency chains is now a standard option โ€” albeit at a premium that’s now transparently priced via the AI Credits rate card. Enterprise and Business admins must explicitly enable it via policy.

VS Code 1.122 & 1.123 โ€” Agent Experience Overhaul

Two VS Code releases landed this week, each carrying significant Copilot agent improvements.

VS Code 1.122 shipped the Agents Window (Preview), a dedicated companion pane for managing agent sessions across projects. BYOK (Bring Your Own Key) now works without a GitHub sign-in, enabling fully offline and air-gapped workflows with local models like Ollama. The Custom Endpoint provider graduated to Stable, 1M token context windows arrived for Claude Opus 4.7 and GPT-5.5, and the model picker now shows per-model credit costs to help users make cost-aware choices.

VS Code 1.123 went further with a specialized execution subagent powered by Google’s Gemini 3 Flash for running shell commands, a revamped Agents window with grid layouts and threaded feedback, and parity between Copilot Cloud tasks and local agent operations. Copilot Chat can now export traces, metrics, and events via OpenTelemetry for real-time agent interaction visibility โ€” a significant step for observability-minded organizations.

For the first time, you can send a Chat request with only attachments and no text, and 1M+ token contexts are supported for selected Anthropic and OpenAI models.

Why This Matters

The VS Code agent experience has matured rapidly. The combination of BYOK without sign-in, Custom Endpoint providers, and OpenTelemetry tracing means organizations with strict compliance requirements (air-gapped networks, self-hosted models, audit trails) can now deploy Copilot agents without compromising their security posture. This is the kind of infrastructure investment that signals GitHub is serious about enterprise adoption.

Agentic Workflows v0.77.4 โ€” Biggest Release Yet

GitHub Agentic Workflows shipped v0.77.4 on May 31, one of the largest single releases in the project’s history. Key additions include:

  • Anthropic Workload Identity Federation (WIF) authentication โ€” eliminates long-lived API keys from repositories by authenticating through GitHub’s identity infrastructure
  • copilot-sdk engine โ€” a new engine: copilot-sdk frontmatter option giving workflows programmatic access to the Copilot SDK runtime
  • Per-workflow 24-hour effective-token guardrails โ€” enterprise-grade cost controls with ET shorthand
  • Modular workflow composition โ€” new includes, skills, and agents keys in aw.yml for composing workflows across repositories
  • GitHub MCP Search Toolset โ€” expanded with search_commits capability
  • New skills โ€” copilot-review for PR review feedback and go-codemod for Go codemod operations

The v0.77.3 release (May 29) added custom auth headers for internal service auth, auto-scaffolding of custom agents via gh aw init, and stricter compile-time schema validation.

CLI Ecosystem Accelerates

The Copilot CLI ecosystem is rapidly maturing:

  • Copilot CLI v1.0.57 (June 1) introduces plugin slash commands โ€” /plugin install, /plugin uninstall, /plugin list โ€” alongside better error messages for rate-limit scenarios
  • The SonarQube plugin brings deterministic static analysis to terminal agent workflows via /sonarqube slash commands
  • A community marketplace for CLI plugins is emerging, with plugins supporting MCP servers, custom skills, and BYOK models
  • The /fleet command enables parallel multi-file operations; /chronicle provides session history insights
  • Copilot Memory gained a repository-level off switch and CLI management commands (/memory on|off|show)
  • Remote control of CLI sessions is now GA โ€” steer agent sessions from github.com or GitHub Mobile

In a telling signal, Microsoft reportedly began transitioning internal Experiences + Devices engineers from Claude Code to GitHub Copilot CLI, targeting completion by June 2026.

Copilot Usage Metrics โ€” Adoption Cohorts

The Copilot usage metrics API now classifies users into AI adoption phases (Phase 0โ€“3) based on interaction patterns. This moves measurement beyond simple active/inactive tracking to understanding how deeply teams are integrating Copilot into their workflows โ€” a useful signal for org-wide rollout tracking and ROI analysis.

Model & Extensions Landscape

Several notable changes hit the model ecosystem this week:

  • Targeted model rules (public preview) let enterprise owners control which models are available per organization, replacing the single enterprise-wide setting
  • Evaluation models are now served in auto-selection for individual plans, giving users access to newer models before they become defaults
  • GitHub removed all Gemini models plus GPT-5.2 Codex and GPT-5.4 nano from Copilot Chat on the web, citing reliability and simplification โ€” though VS Code and JetBrains surfaces may still offer them
  • The GitHub MCP Server v1.1.2 added lockdown mode, issue fields support, commit search, and GHAS alert pagination
  • GitHub published official documentation on MCP server integration with the Copilot SDK, formalizing the extensibility pattern

What to Watch

  • New sign-ups for Pro, Pro+, and Max remain paused โ€” GitHub promises they’ll reopen “in the coming weeks.” Teams currently on the sidelines should have onboarding plans ready.
  • The Copilot App (native desktop) remains in technical preview, offering parallel agent sessions with isolated git worktrees โ€” a potential game-changer for heavy multitaskers.
  • Coding agent penetration is now measurable โ€” a study published May 27 found AI coding agents actively used in 22โ€“29% of 128,018 GitHub projects analyzed, confirming agent usage is mainstream, not experimental.
  • Watch for cost optimization tooling โ€” the billing backlash will almost certainly drive new community and first-party tools for monitoring and optimizing AI Credit consumption.

That’s your Copilot Weekly for June 2, 2026. The billing transition makes this a defining week for the platform โ€” the economics of agentic development just changed in a big way. We’ll be watching how teams adapt and what GitHub does to address the cost concerns. Check back next week.