Glossary
What is GitHub Copilot?
In short
GitHub Copilot is an AI helper for programmers from GitHub that suggests code, answers questions about a project and does tasks as an agent.
Models and products · GitHub Copilot
MMOOZZGG.com editorsChecked: 1 min read
For a child
Imagine you are building a complex castle from blocks by a plan, and next to you sits a friend who has built many castles. While you place a block, the friend tells you which one to put next. Sometimes the friend is wrong, and the tower comes out crooked. So think about every hint instead of placing a block without looking.
In plain words
It is a helper that works inside a developer’s tools. While a programmer types, it suggests the rest of the line or a whole function, and the programmer can accept or reject the hint. In chat you ask it about files and the project, and in agent mode you give it a whole goal: for example, fix a bug from a task description. The helper then makes the changes itself and offers them for review. It also reviews colleagues’ changes and writes comments on them. A newcomer to the team can ask it where the project handles payments.
For business
GitHub’s trade controls rules name Russia among the destinations where the sale, export and re-export of Copilot are restricted, together with Cuba, Iran, North Korea and Belarus (checked 16 October 2026). So a Russian company cannot officially buy the Business or Enterprise plans. Besides access, assess the risks of the code itself: check suggestions as strictly as a junior developer’s work, or errors and vulnerabilities will reach the product. For teams that need a coding helper inside their own network, the answer is an open code model deployed on their own servers, which keeps them independent of a single vendor.
For a specialist
The first versions ran on the Codex model: the paper by Chen et al., “Evaluating Large Language Models Trained on Code” (July 2021), says a separate production version of Codex powers GitHub Copilot. The same paper introduced the HumanEval set: Codex solved 28.8% of the tasks, GPT-3 solved none, and with 100 attempts per task Codex reached 70.2%. According to GitHub documentation as of 16 October 2026, the product includes code completions as you type, chat about code and the repository, an agent mode for multi-step tasks, an agent that takes an issue and returns a pull request, and code review of pull requests. There is a free plan, paid individual plans, Business and Enterprise; the feature set depends on the plan, the client and the organization’s policies.
Example
A developer opens an unfamiliar module and asks the helper to explain how refunds are processed. The developer runs the suggested fix through tests and gives it to a colleague for review before merging it into the main branch.
Related terms
Questions and answers
Sources
Not sure whether your company needs this?
AI solutions map