The short answer
GitHub Copilot is the AI assistant for programmers from GitHub, the platform where most developers keep their code. Copilot suggests code as you type, answers questions about your code in a chat panel, and can work as an agent that carries out a task on its own: editing files, running tests and preparing a pull request.
GitHub released Copilot in 2021, the first AI coding tool used at scale. The first version ran on Codex, a language model from OpenAI. That model has been retired, and Copilot now works with models from several providers. GitHub has been owned by Microsoft since 2018.
Mind the name. Microsoft uses “Copilot” for many products: the assistant in Word and Outlook is Microsoft 365 Copilot, and Copilot Studio is where you build your own agents. GitHub Copilot is the version for writing software.
What GitHub Copilot does
Copilot works at several levels, from small to large. The bigger the task, the more the AI does on its own and the more your review matters.
Completion as you type
Copilot predicts the next lines and shows them greyed out in your editor. Press Tab to accept. This is where Copilot started, and it works best on predictable code.
Chat about your code
In a chat panel next to your code you ask questions: what does this function do, why does this test fail, how do I write this shorter. Copilot looks at the files you have open.
Agent in your editor
In agent mode you give an instruction in plain language. Copilot finds the files involved, edits them, runs commands and fixes what goes wrong. You approve the steps.
Agent on GitHub
You assign an issue to Copilot. It works the task in an isolated environment on GitHub and delivers the result as a pull request for you to review.
Reviewing pull requests
Copilot can read through a pull request and leave comments on what it notices, such as a bug or a missed edge case.
Copilot works in VS Code, Visual Studio, the JetBrains editors, Xcode and a few others, on the GitHub website and in the terminal. You put your project’s conventions in an instructions file that Copilot reads on every task. That combination of model, tools and conventions is called an AI harness; read what is an AI harness for the details.
How much does GitHub Copilot cost?
There is a free version with a monthly cap on completions and chat messages. Enough to try it, too little for daily work. Above it are paid plans for individual developers and separate plans for teams and companies, with management features and terms about your code. Students, teachers and maintainers of popular open source projects can use Copilot for free.
Every plan includes a number of heavier requests, for the agent and the most capable models. If a lot of work runs through those, you pay extra or wait until next month. Current prices and limits are on GitHub’s website.
Copilot, Codex, Claude Code and Cursor side by side
Copilot sits between two kinds of tools. Like Cursor, it helps while you type yourself, and like Codex and Claude Code, it can carry out whole tasks on its own.
| GitHub Copilot | Codex | Claude Code | Cursor | |
|---|---|---|---|---|
| Made by | GitHub, part of Microsoft. | OpenAI, the company behind ChatGPT. | Anthropic, the company behind Claude. | Anysphere, part of SpaceX since 2026. |
| What it is | An assistant in your editor that completes, chats and carries out tasks as an agent. | A coding agent that carries out tasks on its own, locally or in the cloud. | A coding agent that carries out tasks on its own in your project. | A code editor with AI built in, based on VS Code. |
| Where you work | In the editor you already use, on GitHub and in the terminal. | Terminal, editor, ChatGPT in the browser or the desktop app. | Terminal, VS Code, JetBrains, desktop app or browser. | In the editor itself. |
| Models | A choice of models from OpenAI, Anthropic and Google, among others. | OpenAI's models. | Anthropic's Claude models. | A choice of models from several providers. |
| Strong at | Working in your existing editor, and handing issues off on GitHub. | Several tasks at once in the cloud. | Carrying out large tasks on its own in the terminal. | Typing yourself with fast, smart completion. |
| Plan | A free version, with paid plans above it. | Included in ChatGPT plans, or pay per use through the API. | Included in paid Claude plans, or pay per use through the API. | Its own subscription, with a free version. |
Copilot’s biggest advantage is that you switch nothing: it works in your own editor and in the GitHub environment where your code already lives. If you mostly want an agent that finishes whole tasks by itself, Claude Code and Codex go further. If you want to type yourself with the best completion, look at Cursor too. The difference between the tools is smaller than the difference a well set-up project makes.
How a senior team works with Copilot
Whichever tool you pick, the way of working that gets good results is the same. The difference lies in how you give the instruction and what you do with the outcome.
An instructions file in the project
Put the test commands, the folder structure, the conventions and what is off limits in the instructions file. Every mistake Copilot makes twice becomes a rule in that file.
Choosing the mode per task
Completion for code you write yourself, the agent for a well-defined task: a bug, a test, a screen. A vague instruction gets a vague result, in every mode.
Tests the agent runs itself
An agent that can test its own work finds its own mistakes. Make sure the tests also run in the GitHub environment where the agent works.
Actually reading completions
Pressing Tab is faster than reading. Completions are exactly where mistakes slip in unnoticed: a wrong variable, a missing edge case.
Reviewing every pull request
What Copilot produces goes through review like a colleague's work. On security, payments and customer data, a senior developer looks extra hard.
Accepting a suggestion takes one keystroke. Knowing whether it is right is the actual work.
That is the limit of what Copilot delivers: how fast someone with experience can judge what comes back. For how that looks in a full project, from first setup to production, read how to develop software with AI.
Where GitHub Copilot is less suitable
Copilot is made for people who can read code. Without that knowledge you cannot judge whether a suggestion is right, and small mistakes pile up unnoticed. If you want to build an app without programming knowledge, start with a tool like Lovable.
For large tasks that span many files, specialised agents like Claude Code and Codex often get further without steering. And on a large, old codebase without tests, every AI tool needs a lot of guidance: sort out the tests and the instructions file first. Which tool fits which situation is in the best AI harness for coding.
Conclusion: the AI assistant already in your editor
For many teams GitHub Copilot is the first AI tool, because it works in the editor and on the platform they already use. It completes code, answers questions and carries out small tasks as an agent. What produces good work is the same as with any AI tool: a clearly set-up project, well-defined tasks and a review by someone who knows where things can go wrong.
Copilot = writing along and executing
Completing as you type, answering questions and finishing small tasks as an agent.
Senior developer = judging
The instruction, the choices and the review. That is where quality lives, and you don't leave it to the AI.
What is GitHub Copilot?
Is GitHub Copilot free?
What is the difference between GitHub Copilot and Microsoft Copilot?
What is the difference between GitHub Copilot and ChatGPT?
What is the difference between GitHub Copilot and Cursor?
Does GitHub Copilot use my code to train AI?
Build faster with AI without giving up quality?
Our senior developers work with AI tools like Copilot and coding agents every day. Bring them onto your project, or have your team learn in a workshop how to do it well themselves.