AI agents join the coding: 40 percent of GitHub contributions now start with a machine
At its Universe conference, GitHub introduced Copilot Fusion, which picks the right AI model for each step on its own. Its new annual statistics show how strongly agents are changing programming.
Graphic: TERRA WORLD NEWS (own graphic based on the sources listed)
Video: YouTube · GitHub · youtube.com
Microsoft subsidiary GitHub on Wednesday introduced Copilot Fusion at its Universe developer conference. For every step of a coding task – reasoning, writing code, debugging, using tools – the system decides which AI model, or combination of models, handles it. According to Jay Parikh, who leads Microsoft's CoreAI division, the strongest version, Fusion Max, scores within two points of Claude Opus 5.5 Max on Artificial Analysis' Coding Agent Index, at half the cost and with 40 percent less time per task. The video shows the opening keynote.
At the same time, GitHub published its annual Octoverse report for the twelve months to the end of September. It found that in September more than 40 percent of all pull requests – proposals for code changes – started as a draft by an AI agent, up from 17 percent in March. In September alone there were 110 million of them. GitHub stresses that it cannot detect every agent, so the figure is a floor.
In total, the platform counts more than 225 million developers, 34.3 billion commits and 1.3 billion merged pull requests. At 17.9 billion, the number of code pushes exceeded 2024 and 2025 combined. Germany has 6.3 million developers. Enterprise customers on GitHub Enterprise can now – for the time being through GitHub's sales team – use all Copilot features without a separate subscription and pay only for the AI credits they use.