GitHub Copilot Guide

GitHub Copilot is the most widely adopted AI coding tool, integrated directly into VS Code, JetBrains, and Neovim.

Key Features

Inline Completions

The original feature — predictive code suggestions as you type. Uses context from open files and your repository.

Copilot Chat

Conversational AI assistant in the IDE sidebar. Ask questions about code, generate functions, explain errors.

Copilot Workspace (Agents)

GitHub's agentic feature — takes an issue and generates a multi-file implementation plan, then executes it.

1. Start from a GitHub Issue
2. Copilot analyzes the issue and proposes a plan
3. Review and refine the plan
4. Copilot generates code across multiple files
5. Review, test, and create a PR

Copilot CLI

The standalone GitHub Copilot CLI replaced the older gh copilot extension (now retired). Install and invocation follow the current GitHub Copilot CLI docs — check there for the exact install command; the older gh copilot explain / gh copilot suggest subcommands are deprecated.

Pricing

As of June 1 2026, Copilot moved from "premium request" quotas to usage-based GitHub AI Credits (1 credit = $0.01 USD), billed on token consumption across chat, agent mode, code review, and CLI; code completions stay unlimited on paid tiers. See GitHub Copilot plans for current details.

Best For

• Teams already on GitHub — zero friction, native integration
• Enterprise environments needing compliance + audit trails
• Developers who want AI assistance without changing their editor

Gotchas

Based on general developer reports; cross-check against GitHub Copilot docs for the authoritative current state.

• Context is mostly open files. Copilot's workspace index improves this on recent tiers, but Claude Code and Cursor typically see more of a large repo at once.
• AI Credits burn fast. Agent / tool-using tasks consume usage-based AI Credits (token-metered since June 2026) faster than most users expect. A heavy Pro user can exhaust the $15/mo included credits well before month-end and start paying overage.
• Per-seat billing. Business + Enterprise are seat-based. A small team with one heavy user and several light users still pays for every seat.
• Limited bring-your-own-model. Model selection is growing but narrower than Cursor or Windsurf. No Anthropic API key passthrough.
• Latency on tool-using tasks. Agent actions involve an IDE round-trip that adds perceptible delay vs terminal-native tools like Claude Code.
• Context files path. Uses .github/copilot-instructions.md (repo-scoped); does not read CLAUDE.md or .cursorrules.