Particle.news
Download on the App Store

GitHub Recasts Copilot as an Agentic Software Delivery System

The shift centers Copilot Fusion as a model router, introduces new agent products, promises large infrastructure scaling to handle rising AI-driven commit volumes.

Overview

  • GitHub announced on Oct. 28 that it has repositioned Copilot from an autocomplete assistant into a platform that orchestrates AI agents, tools, and human decisions across the software development lifecycle.
  • Copilot Fusion is now generally available and automatically selects and routes tasks across models with effort-level presets called Standard, High, and Max.
  • GitHub reported company benchmarks saying Fusion Max scored within two points of Opus 5.5 Max on the Artificial Analysis index while using about half the API cost and about 40% less time per task, and those claims remain company-reported and unverified externally.
  • Octoverse data presented at the event showed agents opened more than 40% of pull requests in September, a surge that produced roughly 110 million PRs and 7.4 billion commits in the month and which GitHub said stressed its platform and helped cause earlier outages.
  • To address scale GitHub added capacity (4 million CPU cores, 160 PB storage, 50 Tb/s network), announced a planned multi-region Azure migration and a Git redesign, and moved enterprise Copilot to a credits-based billing model that charges per AI task with admin controls over allowed models.