GitHub Copilot

Quick Verdict

GitHub Copilot is no longer only an autocomplete extension. The current product covers inline suggestions, editor chat and agent mode, GitHub pull-request review, a command-line agent, cloud coding agents, MCP connections, and delegation to third-party agents. The “Copilot X” label announced in 2023 is historical; these capabilities now belong to the main GitHub Copilot product.

Copilot remains a strong default for developers who want AI inside existing IDE and GitHub workflows. It can stay lightweight, offering completions and explanations, or become more autonomous by changing files, running tools, and creating pull requests from issues. That expansion requires stricter controls. An agent that can read a repository and run commands must receive least privilege, cannot be given plaintext production secrets, and should work behind protected branches, required reviews, and CI.

Cost also changed in June 2026. Chat, agent mode, code review, cloud agents, CLI, Spaces, and Spark consume GitHub AI Credits according to the model and task. Paid plans keep code completion and next-edit suggestions unlimited without credit consumption. Long agent sessions, premium models, repeated attempts, and code review can therefore make actual spend differ substantially between developers.

Best For

• Individual developers who want completion, chat, and agents without abandoning a familiar editor.
• Teams already using GitHub Issues, pull requests, Actions, CODEOWNERS, and protected branches.
• Organizations that need centralized model, MCP, agent, access, budget, and audit policies.
• Students and occasional users who can begin with limited Free usage before paying.
• It is not suitable as an unreviewed deployment system or for repositories without accountable maintainers and automated tests.

Key Features

• Completion and next edits: Provides inline suggestions in VS Code, Visual Studio, JetBrains, Xcode, Neovim, Eclipse, and other supported environments. These remain unlimited and credit-free on paid plans.
• Unified Visual Studio component: Visual Studio 2022 version 17.10 and later installs and updates completion and chat through one optional GitHub Copilot component in Visual Studio Installer. Version 17.9 and earlier continue to manage the extension through Manage Extensions.
• IntelliSense and IntelliCode separation: IntelliSense is the IDE’s language-service completion; IntelliCode is bundled, free AI-assisted ranking and supported local whole-line suggestions; Copilot adds separately entitled cloud generation, chat, and agents.
• Chat and agent mode: Explains code, proposes tests, changes multiple files, and runs tools in supported editors. Models and capabilities vary by plan and environment.
• Cloud agent: Accepts delegated work from a GitHub issue or integrated surface, researches the repository, implements a change, and opens a pull request.
• Code review: Reviews pull-request or editor diffs and suggests issues. It is a supplemental reviewer, not a merge authority or a replacement for business knowledge.
• Copilot CLI: Explains and performs terminal development work. Users must inspect generated shell commands for deletion, networking, privilege, and exfiltration risk before execution.
• Models, MCP, and third-party agents: Supports model choice, controlled MCP servers, custom agents, and delegation to agents such as Claude Code and OpenAI Codex, depending on the plan.
• Organization controls: Business and Enterprise add access and budget management, governance, privacy commitments, and other organization capabilities.

Use Cases

For normal development, begin with completion and bounded chat. Select the relevant code, describe the expected behavior, and ask for a minimal change rather than a broad rewrite. When generating tests, state business invariants, boundaries, and failure paths first; otherwise the model may merely mirror the implementation and preserve its bug.

Agent mode needs an explicit scope: allowed directories, files that must not change, test commands, acceptance criteria, and whether network access is permitted. Review the workspace diff before accepting commands. Cloud agents and third-party agents should use dedicated branches with access only to named repositories. Preserve required reviews, CODEOWNERS, status checks, and restrictions on direct pushes to the default branch.

Never paste API keys, production configuration, customer records, or .env contents into chat, issues, or prompts. Use protected GitHub Actions secrets, environment rules, isolated test services, and short-lived credentials. MCP servers need an allow list and a separate review of every exposed read or write tool. A harmless-looking documentation server and a production database server do not have the same blast radius.

Every generated change needs a complete diff review. Check dependencies, licenses, authentication, input validation, error paths, logging, migrations, concurrency, and rollback. Run formatting, static analysis, type checks, unit and integration tests, security scanning, and relevant end-to-end suites. Copilot code review may identify additional defects, but the reviewer still needs system and business context.

For AI-native editor workflows, compare Cursor and Windsurf. For direct terminal control, compare Claude Code. For longer independently delegated assignments, Devin is another benchmark.

Pricing

Visual Studio host licensing and Copilot entitlement are separate decisions. Installing the unified Copilot component in Visual Studio 17.10+ does not grant a Professional or Enterprise IDE license, and it does not by itself grant Copilot access. A user still needs eligible Community use or a commercial Visual Studio license, plus a GitHub account entitled through Copilot Free, an individual paid plan, or an organization-assigned seat. IntelliSense and the IntelliCode features bundled with Visual Studio do not require a Copilot seat.

The following public individual prices and included allowances were shown by GitHub on July 21, 2026. One GitHub AI Credit equals $0.01. GitHub can change flex allotments, models, and features, so checkout, taxes, overage, annual-plan transitions, and organization contracts must be rechecked.

Paid completions and next-edit suggestions do not consume credits. Chat, agents, CLI, Spaces, Spark, and review do, with cost depending on model and complexity. From June 1, 2026, code-review workflows also consume GitHub Actions minutes. Administrators should set a paid-usage policy, dollar budgets, and 75%, 90%, and 100% alerts. Analyze spending by developer, repository, feature, and accepted pull request instead of assuming the seat price is the total cost.

Pros

• Covers editor, terminal, GitHub, pull-request review, and cloud-agent workflows in one product.
• Reuses GitHub repositories, issues, pull requests, Actions, and organization controls.
• Offers model choice, MCP, custom agents, and third-party agent delegation.
• Provides a progression from limited Free use to individual high-volume and organization plans.
• GitHub states that Business and Enterprise data is not used to train its models.

Cons

• AI Credits make agent, chat, review, and premium-model costs less predictable than old fixed request quotas.
• Cloud agents, CLI, MCP, and third-party agents widen repository, command, and external-system permissions.
• GitHub may use interactions from individual Free, Pro, and Pro+ users for model improvement unless they opt out.
• Code review can consume both AI Credits and Actions minutes.
• Generated changes can still contain logic, security, dependency, license, and outdated-API problems.
• GitHub, model endpoints, authentication, and cloud agents may be unreliable from mainland China, so this directory marks needsVPN: true.

Alternatives

FAQ

What are the current GitHub Copilot plans?

Individuals can choose Free, Pro, Pro+, or Max. Organizations can choose Business or Enterprise. Features, models, included credits, and prices change, so verify the official plans page and checkout.

Do code completions consume GitHub AI Credits?

Paid-plan code completions and next-edit suggestions do not consume credits. Chat, agent mode, cloud agents, code review, CLI, Spaces, and Spark consume credits based on model and task complexity.

Can Copilot access an entire private repository?

Access depends on the account, installation, feature, and organization policies. Grant only required repositories, restrict MCP and third-party agents, keep content exclusion and audit controls, and avoid routine work with administrator identities.

How should API keys and production secrets be protected?

Do not paste them into chat, issues, prompts, or code. Use Actions secrets, protected environments, and short-lived test credentials. Run secret scanning on commits and revoke and rotate any exposed credential immediately.

Can Copilot-generated or Copilot-reviewed code be merged directly?

No. A human must inspect the complete diff, dependencies, licenses, permissions, and migrations, then run type, unit, integration, security, and applicable end-to-end tests through the normal pull-request process.

Does GitHub use Copilot data for training?

GitHub says individual Free, Pro, and Pro+ interactions may be used to improve models unless the user opts out in settings. Business and Enterprise data is not used for training. Review current terms and retention rules before procurement.

Bottom Line

GitHub Copilot now connects completion, editor and terminal agents, pull-request review, and cloud delegation under GitHub’s development controls. It is a strong fit for teams that already have engineering gates and want to introduce autonomy gradually. Start with completion and small agent tasks, enforce least privilege, isolate secrets, require human review and comprehensive tests, and set AI Credit budgets. Expand cloud and third-party agents only after quality and accepted-PR cost remain predictable.