AI 101

GitHub Copilot: the agent inside the editor you already use

Install Copilot in VS Code, sign in with a GitHub account (free plan included), switch Chat into agent mode, and use the Copilot CLI for terminal work. Plans and AI credits explained without the marketing.

By the end of this page, you'll have Copilot running in VS Code with agent mode enabled, you'll understand what a "GitHub AI credit" is and which plan you actually need, and you'll have the Copilot CLI installed for the times a terminal is the right surface.

What Copilot is now

Copilot started in 2021 as autocomplete with opinions. In 2026 it’s three products sharing one subscription:

  1. Completions โ€” ghost text in the editor, still the thing you’ll use every minute.
  2. Chat + agent mode โ€” a panel that can read your workspace, edit multiple files, run terminal commands, and call MCP tools, with your approval.
  3. Copilot CLI โ€” the same agent in a terminal, for repos and scripts outside the editor.

The pricing model changed with it: Copilot is now metered in GitHub AI credits rather than “chat requests”. Each plan includes a monthly allowance; heavy agent use burns through it.

Step 1 โ€” Install in VS Code

  1. Open VS Code โ†’ Extensions view.
  2. Install GitHub Copilot (Microsoft) โ€” the Chat extension comes with it.
  3. Click the Copilot icon in the status bar โ†’ Sign in with GitHub.
  4. Authorize in the browser, come back to VS Code.

No subscription? The Free plan activates with your GitHub account. It’s genuinely usable for learning: completions, chat, and agent mode with a monthly credit allowance and automatic model selection.

JetBrains, Visual Studio, Neovim, Xcode: Copilot ships extensions for all of them. The agent features land in VS Code first; the other IDEs lag by weeks to months.

Step 2 โ€” Switch Chat into agent mode

Open Chat from the Copilot icon, then look at the mode selector at the bottom of the input box:

  • Ask โ€” question/answer, reads context, changes nothing.
  • Edit โ€” multi-file edits you review as a diff.
  • Agent โ€” plans, edits, runs commands, reads output, iterates. This is the one that does work.

In agent mode, every terminal command and file write is proposed first. You approve, skip, or edit. Approval fatigue is the real risk โ€” after the tenth npm test it’s tempting to hit “always allow”. Don’t do that for anything that writes outside the repo.

Two settings worth knowing:

  • Custom instructions: .github/copilot-instructions.md at the repo root is read automatically. Put your conventions there, not in chat.
  • MCP servers: Copilot can call external tools (databases, browsers, issue trackers) through MCP. Start with one, not five. What MCP is โ†’

Step 3 โ€” Copilot CLI

For work outside the editor โ€” a server over SSH, a script folder, a CI job:

npm install -g @github/copilot      # Node.js 22+
winget install GitHub.Copilot       # Windows
brew install --cask copilot-cli     # macOS / Linux

First run:

cd ~/projects/my-app
copilot
/login          # authenticate with GitHub, once

Then just ask. Shortcuts that matter: @ to mention files, / for slash commands, ? for help, Esc to cancel, Ctrl+C to clear or exit.

Non-interactive, for scripts:

copilot -p "explain what this repo does"
copilot -sp "list the TODO comments in src/"   # -s: response only, no usage noise
Trust prompt: the CLI asks you to confirm the current directory is safe to hand to an AI tool. Read it. Running copilot from ~ gives an agent read access to your SSH keys, browser profiles and every .env on the machine.

Step 4 โ€” Which plan, really

Prices as of November 2026, per user per month:

PlanPriceAI credits / monthWho it’s for
Free$0small allowancetrying it out, students
Pro$101,000one developer, daily use
Pro+$393,900heavy agent use, premium models
Max$10010,000all-day agent work
Business$19/seat1,900orgs with policy controls
Enterprise$39/seatlarger poolGitHub Enterprise Cloud

A credit is not a request: a long agent turn with a big model can cost many credits, a completion costs a fraction. The practical rule: start on Free, move to Pro the first month you hit the cap, and only consider Pro+ if you’re in agent mode most of the day.

Copilot vs. Cursor vs. Claude Code

CopilotCursorClaude Code
SurfaceVS Code / JetBrains / CLIits own editorterminal
Free tieryesyes (limited)no
Model choiceGitHub-hosted selectionmulti-vendorAnthropic only
Strengthcheapest path from zero, enterprise controlsbest editor UXbest long agent sessions
Weaknessagent features lagsubscription upheaval (see next page)no free tier

If your team already pays for GitHub, Copilot is the least-friction choice and the one your security team has already reviewed. If you live in a terminal, the CLI is good but Claude Code and Codex CLI are more capable there.

Next in the cloud track: [Cursor โ€” the AI-first editor, now sold through The X Subscription โ†’](/tutorials/cloud/cursor/)