πŸ–₯️

GitHub Copilot App (2026)

Announced at Build 2026, technical preview May 14, generally available June 17, and on every Copilot plan (including Free and Education) since July 7, 2026. The app is the desktop home for agent work on macOS, Windows, and Linux, built on the same Copilot SDK runtime as the CLI and the VS Code Agent Host.

What the App Is

  • The agent-native desktop experience β€” find, run, steer, and land work across repos
  • github/app β€” downloads, release notes, issues
  • My Work: sessions, issues, PRs, and automations across connected repos in one view
  • Sessions: start from an issue, PR, or prompt; each local session runs in its own git worktree, cloud sessions run in GitHub-hosted environments
  • Canvases: plan, PR, terminal, and browser surfaces you and the agent operate on together
  • Automations: repeatable prompts as scheduled or background tasks
  • Agent Merge: the agent follows a PR through CI failures, review comments, and merge under conditions you set

Governance for the App

Customize: Skills, Plugins, MCP

  • Customize tab (GA Aug 25, 2026) β€” MCP servers, plugins, skills, and canvases in one place
  • Agent Plugins 1.0 β€” open spec (Aug 6, 2026) from Amazon, Anysphere, Microsoft, OpenAI, Vercel: plugin.json + skills/ + mcp.json
  • Agent Skills β€” SKILL.md folders in .github/skills, .claude/skills, or .agents/skills
  • BYOK model providers, 1M-token context window, model picker with Low/Medium/High reasoning effort

Copilot SDK & Agent Host

  • Copilot SDK β€” GA June 2, 2026 in Node/TypeScript, Python, Go, .NET, Rust, Java; the runtime behind the app, CLI, and IDE agents
  • VS Code Agent Host Protocol (July 2026) runs Copilot, Claude, and Codex harnesses in a dedicated process; sessions move between the app, CLI, and editor
  • Use the SDK when your team needs an internal agent with the same guardrails (custom tools, MCP, approvals) instead of a bespoke orchestration layer
🧭

Agentic AI & Agent Mode

Copilot Cloud Agent & Agent Mode

Copilot CLI & Terminal Agents

  • Copilot CLI GA (Feb 25, 2026) β€” github/copilot-cli
  • /fleet β€” decomposes a plan into parallel subagents (Apr 2026); Plan mode, Autopilot, --plan --mode autopilot
  • Rubber Duck β€” second opinion from a different model family, on by default
  • /security-review, /worktree sessions, /permissions approval modes, sandboxes with enterprise-managed policies
  • Hooks (preToolUse / postToolUse) with OpenTelemetry trace context, repository memory, /plugin, /mcp, /skills, 200K vs. 1M context tiers
  • Multi-model: Claude Sonnet 5 / Opus 5, GPT-5.6, Gemini 3.6/3.7 Flash, Grok 4.6 (xhigh reasoning), MAI-Code

Custom Agents, Skills & agents.md

  • Creating custom agents β€” file-based .agent.md profiles; org and enterprise owners can publish agents org-wide (Aug 2026)
  • How to write a great agents.md β€” lessons from 2,500+ repos; now read by Copilot code review too
  • awesome-copilot skills β€” reusable SKILL.md examples
  • Scopes: repo (.github/agents, .github/skills), org (.github-private), personal (~/.copilot/agents, ~/.copilot/skills)
  • Package skills + MCP as an Agent Plugin to share across Copilot, Codex, Cursor, Kiro, and ChatGPT

Mission Control & Agent HQ

  • Introducing Agent HQ β€” any agent, any way you work
  • How to orchestrate agents using mission control β€” assign across repos, pick a custom agent, watch session logs, steer mid-run
  • Mission Control on github.com is the web command center; the Copilot app is its desktop counterpart with local worktrees and canvases
  • Model selection for Claude and Codex agents (Apr 2026); third-party agents from Anthropic, OpenAI, Google, Cognition, xAI
  • Agent identity & access management (treat agents like team members); integrations: Slack, Linear, Jira, Teams, Azure Boards, Raycast
πŸ›‘οΈ

MCP (Model Context Protocol)

Registry & Guides

Security & Governance

Enterprise MCP Controls

πŸ“ˆ

Developer Experience & Workflows

Assessment Tools

Health Checks

Agentic Code Review

  • Agentic tool-calling architecture GA (Mar 5, 2026) β€” gathers repo context as needed
  • AGENTS.md support; also reads REVIEW.md, CLAUDE.md, GEMINI.md, and instructions from the PR head branch
  • Effort levels GA (Aug 2026); runs behind a firewall by default, configurable separately from cloud agent
  • Cloud agent self-reviews with Copilot code review before opening a PR; add Rubber Duck for a cross-family second opinion
  • Risk-based review tiers: human-authored vs. agent-authored PRs (see code quality gates)

VS Code Agent Host (Aug 2026)

  • August 2026 releases (v1.132–1.135) β€” external agent sessions from the app or CLI continue in VS Code
  • VS Code v1.135 β€” Rubber Duck (experimental), side-by-side chats, /btw side conversations, prompt timeline, browser element comments
  • Agent Host Protocol (July 2026): Copilot, Claude, and Codex harnesses in a dedicated process; Copilot agent powered by the Copilot SDK
  • AI credit usage in the status menu, BYOK models in the Agents window, ! terminal commands from chat
  • v1.109 (Jan 2026) β€” Copilot Memory, terminal sandboxing, multi-agent hub

Models & Metrics (Sept 2026)

🌐

The Ecosystem

Tools & Repos

Agent Context System

  • Agent Context System β€” persistent context, multi-tool orchestration, 24/7 agent execution
  • Highlights: context engineering, scoped permissions, autonomous task runners
πŸ“°

Main Branch Newsletter

🎀

Talks & Appearances

  • GitHub Universe 2025 β€” panel with Justin Cormack (Docker) + Quincy Larson (freeCodeCamp)
  • Commit Your Code Conference 2025 β€” hybrid human/AI workflows
  • JFrog SwampUP 2025 β€” AI agent security risks and autonomy
  • GitHub Universe LATAM Series β€” Universe announcements (Spanish)
  • freeCodeCamp Podcast #199 β€” Tips from a Serial Career Changer
  • RedMonk Conversation (Feb 2025) β€” onboarding beginners to Git with Kedasha Kerr
πŸ“Š

Research & Data

  • Deloitte: 25% of companies running agentic pilots in 2025; 50% by 2027.
  • McKinsey: governance-first teams see +37% adoption success, -41% security issues.
  • Microsoft: productivity gains arrive after ~11 weeks of steady use.
  • GitHub Enterprise Research: structured sharing drives ~40% better outcomes.
  • Octoverse 2025: 1.13M repos importing LLM SDKs (+178% YoY); Universe 2025 launched Agent HQ, Mission Control, Custom Copilot Agents.
  • Agent HQ multi-model fleet: Anthropic, OpenAI, Google, Cognition, xAI agents on one platform via Copilot subscription.
  • Copilot app on every plan since July 2026, including Free and Education; Copilot CLI, cloud agent, and the app share one SDK runtime.
  • 2,500+ repos with agents.md analyzed β€” custom agents and skills becoming standard practice; Agent Plugins 1.0 (Aug 2026) shipped with support from ChatGPT, Codex, Cursor, GitHub Copilot, Kiro, and VS Code.
  • Stack Overflow surveys: 84% of developers use or plan to use AI tools, 51% daily, but only ~29% trust output accuracy (down from 40%) β€” the trust gap is the adoption problem to manage.
  • GitHub Rubber Duck evaluation: Claude Sonnet plus a cross-family reviewer closes ~75% of the gap to Opus alone on hard multi-file tasks β€” verification beats bigger models for many teams.
  • Cloud agent retrieval: 2Γ— throughput and 37.6% better retrieval accuracy vs. early 2025; 1M-token context in VS Code, CLI, and the app.
  • AIDev (arXiv, Feb 2026): first large-scale study of AI coding agent PRs on GitHub β€” agent PRs need the same review discipline as human PRs, with tighter scope.