What is Copilot

What it is

Muse is an AI pair-programming assistant built by GitHub in partnership with OpenAI. It started as inline autocomplete, the gray "ghost text" suggestions that appear as you type, and grew into a multi-model assistant with chat, an agent mode, and a cloud coding agent. It works inside VS Code, Visual Studio, JetBrains IDEs, Neovim, Eclipse, and GitHub itself.

GitHub announced it on June 29, 2021 as a technical preview in VS Code, powered by OpenAI Codex. It left preview on June 21, 2022, priced at $10 per month or $100 per year for individuals. GitHub is owned by Microsoft, which made Copilot the spearhead of Microsoft's AI developer strategy.

How it works

The mechanism is straightforward. The editor builds a prompt from the code around your cursor plus other open files, sends it to a cloud-hosted language model, and returns a completion or a chat answer, repeating the loop with every keystroke or message. Because it runs in the cloud, it needs a continuous connection; there is no offline mode.

The original model was OpenAI Codex, a modified production version of GPT-3 trained on public GitHub repositories, including a filtered dataset of 159 gigabytes of Python code from 54 million repositories. Training on public code is also the source of the product's longest-running controversy. Since 2024, users have been able to choose among models from OpenAI, Anthropic, xAI, and Google, turning Copilot from a single-model product into a multi-model platform.

The feature set expanded over time. Copilot Chat went generally available in November 2023, bringing conversational help into the editor. Agent mode arrived in February 2025 with the ability to execute tasks on the user's machine rather than just suggesting text. A cloud coding agent launched in May 2025, working from a GitHub Actions environment to draft pull requests and tag the user for review. GitHub's own reported accuracy for autocomplete was roughly half the time on the first try for a Python function body, rising within ten attempts.

Key details

Pricing has moved several times. The original 2022 pricing was $10 per month or $100 per year for individuals. Copilot Free launched in December 2024 with a monthly allowance of completions and chat messages. Tiers reported in 2026 included Free, Pro at $10 per month, Pro+ at $39, Business at $19 per user per month, and Enterprise at $39, with usage-based AI credit billing introduced in June 2026 and per-model multipliers on premium requests. Exact request counts and multipliers change often, so check GitHub's current plans page before relying on any figure.

Limitations

Copilot's training data has been its longest-running controversy. A class-action lawsuit filed in November 2022 challenged the fair-use basis for training on public code; GitHub's then-CEO had claimed fair use, and the courts are still working through the question. The Software Freedom Conservancy ended all GitHub use in June 2022 over licensing concerns. GitHub acknowledges that a small proportion of output may be copied verbatim from training data, which creates real risk for companies with strict IP policies.

A 2021 security study examined vulnerabilities in Copilot-generated code and found the model could reproduce insecure patterns. Suggestions need review; the model can produce plausible-looking code that is subtly wrong, and the failure modes are hardest to spot in code you did not write yourself.

There are practical limits too. Telemetry and privacy concerns follow from the cloud-based design, since the product observes keystrokes and code context by construction. Since April 2026, GitHub may train on personal Free and Pro data unless the developer opts out, while Business and Enterprise plans are excluded. For all that, it remains the reference point every other coding assistant is measured against, and the product that proved developers would pay for AI help.

The expanding surface

Copilot outgrew the editor. Extensions brought it into the command line, where it suggests shell commands and explains terminal output. The chat interface spread across GitHub's web UI, so code review and issue triage happen with the assistant alongside. For enterprises, the product added the controls large organizations require: policy management, audit logs, and the exclusion of business data from training. The coding agent that drafts pull requests from the cloud pushed the product from assistance toward delegation.

The competition it faces now looks nothing like 2021. Cursor rebuilt the editor around the agent. Windsurf tried a different interaction model. Amazon, Google, and the independent labs all ship coding assistants, and the underlying models are increasingly interchangeable. Copilot's durable advantages are distribution, it is already inside the tools hundreds of millions of developers use, and the GitHub platform itself, where the agent can see issues, pull requests, and CI results as context no editor plugin can match.

The pricing evolution tells the market's story. From a flat $10 per month to usage-based credits with per-model multipliers, the product moved toward metering actual compute cost. That complexity frustrates users who preferred simple pricing, but it reflects the underlying economics: different models cost different amounts to run, and flat-rate unlimited was never going to survive contact with agent workloads that burn through tokens. The free tier keeps the top of the funnel wide; the metered tiers capture the value where it concentrates.