Skip to content
The Shape of Intelligence

GitHub Copilot writes code

A GPT-3 descendant trained on public code completes whole functions from a comment inside the editor; programming is the first profession to get an AI colleague.

category
product
significance
4 of 5
people
Nat Friedman, Mark Chen, Wojciech Zaremba
organisations
GitHub, OpenAI, Microsoft

what had to happen · 36 events back to 1943

Every event this one built on, transitively, in order. Direct influences are marked.

GitHub released Copilot as a technical preview on 29 June 2021. Inside Visual Studio Code, as a programmer typed, it proposed the rest of the line, or the whole function, in grey text, and a press of the tab key accepted it. The model behind it was Codex, a GPT-3 fine-tuned on billions of lines of public code from GitHub, described in a paper the following week that reported it solved about 29 percent of a set of programming problems on the first try and 70 percent given a hundred attempts.

It was the first product to put a large language model into the daily work of a profession, and programmers, who could check the output by running it, adopted it faster than anyone. By 2023 GitHub reported over a million paying users and claimed that Copilot wrote nearly half the code in files where it was enabled. The legal questions it raised, about training on licensed code and reproducing it, went to court in 2022 and largely failed.

Copilot is on this timeline as the beginning of the change that ChatGPT made general eighteen months later, and as the origin of the coding agents of 2025 and 2026, which write, run and fix programs on their own and are, by most measures, the most economically consequential use of the technology.

what it led to · 6 events downstream, through 2026

Built on it directly:

  1. 2024Claude learns to use a computerVI
  2. 2025Claude 4 and Claude CodeVII

And, through them, by era:

sources · 2

See this era in the exhibition →Back to the timeline