GitHub Copilot moved suggestions into the editor, not a chat window

An assistant that suggests inside the file

On 29 June 2021, GitHub announced a technical preview of a tool it called an AI pair programmer. The preview announcement describes a system that reads the code and comments already in a file and offers completions ranging from a single line to a whole function, built on a model GitHub says was trained largely on public source code. That framing matters: rather than a chat window a developer opens to ask a question, Copilot was pitched as suggestions appearing directly inside the editor as code is typed, closer in feel to autocomplete than to a conversation.

From a limited preview to a monthly subscription

The 2021 announcement offered a limited number of technical-preview spots rather than open access. Almost exactly a year later, the general availability announcement opened Copilot to any individual developer at a stated price of 10 US dollars a month or 100 US dollars a year, with free access carved out for verified students and for maintainers of popular open-source projects, and business access promised later in 2022. That is a historical price attached to a specific announcement, not a description of what Copilot costs today. The same announcement reports that during the preview, Copilot had accumulated more than 1.2 million users and that, in popular languages such as Python, close to 40 per cent of newly written code in Copilot-assisted files came from its suggestions, a usage figure rather than an accuracy or quality measure.

What the assistant is not told

GitHub's own Copilot documentation, a living reference rather than a dated announcement, sets out what shapes a suggestion: whether repository indexing is enabled, whether a project has a custom-instructions file, and which paths a team has excluded from Copilot's view. None of that appears in the 2021 pitch. The documentation is explicit that suggestions can match publicly available code and that pull requests opened by Copilot need the same review as a human contributor's, not less. A tool that completes a function is not the same as a tool that understands why the function exists, and the review step is where that gap is meant to be caught.

Questions to ask before you adopt it

• Does the assistant index this codebase, or only the open file, and does that change what it gets wrong?
• Who reviews AI-suggested pull requests, and is that review resourced the same as any other contribution?
• Which paths, secrets or licensed code have been excluded from what the assistant can see?

Copilot's own history is a useful caution: a limited preview became a global subscription business within a year, and the seat price quoted in that transition is now a historical detail, not a current one. What has stayed constant across GitHub's own documentation is the instruction to treat a suggestion as a draft that needs the same scrutiny as any other line of code entering the repository.