GitHub Copilot is an AI coding assistant that integrates into code editors such as Visual Studio Code and suggests inline code completions, whole functions, and, in later versions, multi-file edits and chat-based explanations, based on the surrounding code and natural-language comments. Developed by GitHub, a subsidiary of Microsoft (AI), in partnership with OpenAI, it entered technical preview in June 2021 and became generally available as a paid subscription in June 2022.
Copilot's underlying models were originally derived from OpenAI's GPT-3 codebase, fine-tuned specifically for source code on large volumes of public repositories; GitHub and OpenAI later moved to a mix of proprietary and third-party models, including OpenAI's GPT-4 family and, from 2023 onward, Anthropic's Claude models offered as alternatives within Copilot.
History
Copilot launched at a moment when Large language model code generation was still largely a research curiosity, and its release is widely seen as the moment AI-assisted coding moved into everyday professional use. GitHub reported hundreds of thousands of paying subscribers within its first year and, by 2023, stated that Copilot was writing a significant share of code in enabled repositories, though independent verification of such productivity figures has been limited. Subsequent versions added a chat interface ("Copilot Chat"), pull-request summarization, and an agent mode capable of executing multi-step coding tasks with limited supervision, mirroring capabilities that rival tools such as Cursor and Claude Code introduced around the same period.
How it works
Copilot operates by sending the contents of open files, and in later versions broader repository context, to a language model, which returns ranked code suggestions that the developer can accept, reject, or edit. This is a form of Tool use (function calling) integration between an LLM and an external environment (the editor and, in agent mode, the terminal and file system) rather than a standalone chat product. Its suggestions rely on patterns learned during Fine-tuning on code repositories, meaning quality varies by programming language and by how well-represented a given library or pattern was in its Training data.
Reception and legal challenges
Reception has been broadly positive among individual developers, with studies commissioned by GitHub and independent academic groups reporting measurable, though contested, gains in completion speed for certain tasks. Copilot has also faced sustained criticism and litigation. A 2022 class-action lawsuit, later narrowed by a US court, alleged that Copilot was trained on open-source code without complying with the licenses (such as attribution requirements) those projects required, feeding into the broader wave of AI copyright lawsuits and unresolved questions in AI and copyright law about whether training on public code constitutes fair use and whether generated output can reproduce licensed snippets verbatim. Some studies found Copilot occasionally reproduced memorized code fragments, including licensed or insecure code, raising both legal and security concerns that GitHub has partially addressed with filtering features.
Copilot is frequently cited as the opening act of the broader shift toward AI-assisted and, later, largely AI-driven software development sometimes called Vibe coding, a trend that accelerated sharply with the arrival of agentic tools in 2024 and 2025.