Skip to content

github-copilot.md

Latest commit

 

History

History
60 lines (42 loc) · 3.06 KB

github-copilot.md

File metadata and controls

60 lines (42 loc) · 3.06 KB
Raw

GitHub Copilot

ZH EN Home

One-line take: GitHub Copilot now behaves more like an agent platform than a single completion feature.

Quick Read

Item Conclusion
Vendor GitHub / Microsoft
Route Multi-surface agent platform in the VS Code and GitHub ecosystem
Open source No
Best for Teams that want local, CLI, and cloud paths in one flow
Main cost Best value depends heavily on ecosystem alignment
Official website https://code.visualstudio.com/docs/copilot/agents/overview

When To Pick It

  • Your team already lives in VS Code and GitHub.
  • You want Ask, Plan, Agent, CLI, and cloud delegation in a connected workflow.
  • You want issues, PRs, review, and editor execution to line up naturally.

When Not To Pick It

  • You do not work in the VS Code / GitHub ecosystem.
  • You need a strongly open-source or self-hosted path.
  • You only want a tiny single-surface agent.

Capability Shape

Dimension Assessment Notes
Local execution Strong Editor-native workflows are meaningful here
Cloud delegation Strong Good fit for GitHub-centered handoff
Workflow handoff Strong One of the biggest advantages
Permission control Strong Control levels are fairly explicit
Custom agents Medium Useful for role-based entry points

Mid-2026 Updates

June–July 2026 pushed Copilot further toward an agent platform than an autocomplete product:

  • Agent finder (June 17): Copilot discovers the right capability for a task on its own instead of requiring manual MCP/skills/agents configuration.
  • JetBrains AI Assistant (June 30): Copilot Agent became a first-class option in the JetBrains agent picker — it now competes inside other vendors' surfaces.
  • VS Code (June): agentic browser tools GA (navigate, inspect, screenshot, validate web apps in-editor), agent-session organization with cost visibility, and 1M-token context with compatible Anthropic and OpenAI models.
  • Code review (June 18): repository-level AGENTS.md support in Copilot code review.
  • Copilot CLI (July): plan mode now hard-blocks mutating tool calls while planning, plus voice-mode persistence, canvas support, and tighter subagent controls; a /security-review command entered public preview.

Selection takeaway: the platform advantage keeps compounding — Copilot is increasingly the agent layer of the whole GitHub workflow rather than a single assistant.

Operating Cost

Complexity is Medium. Basic use is easy, but mixing local agents, CLI paths, cloud runs, and multiple permission modes makes the workflow more complex fast.

Bottom Line

The main advantage is not one super-agent. It is that mainstream development workflows can live in one surface.