GitHub launched Copilot Fusion at its Universe conference, a routing system that selects the model or mix of models for each step of a coding task. Fusion Max scored within two points of Anthropic Claude Opus 5.5 Max on the Artificial Analysis Coding Agent Index at half the cost and with 40% less time per task. The launch matters because agents already initiated more than 40% of pull requests by September, which shifts coding work from manual writing to supervision.
Copilot Fusion routing and Universe releases
Microsoft CEO Satya Nadella said Fusion delivers frontier-level quality 40% faster and at up to 77% lower cost. The system weighs requirements across reasoning, code generation, debugging and tool use, according to executive vice president Jay Parikh. The unified system behind Copilot agents is now available in Foundry, Microsoft managed service for long-running agents. GitHub also introduced the GitHub Copilot app with Copilot projects that run agent work across repositories in parallel managed sessions.
Session continuity allows one cloud agent session to move between the app, VS Code, the command line and the web. Copilot automations can now learn from results and propose changes for developers to approve. For review, GitHub added Copilot code review Max, which posted the highest F1 score on ReviewBench for high-severity issues. From Wednesday, GitHub Enterprise customers receive all Copilot features without a separate subscription and pay for AI credits used through the sales team, with self-serve coming soon.
Developers pushed code 17.9 billion times in the 12 months to 30 September, more than in 2024 and 2025 combined, according to the annual Octoverse report. Commits reached 34.3 billion and merged pull requests reached 1.3 billion. The platform now counts more than 225 million developers, including 38.3 million in the United States and 32 million in India. TypeScript remained the most used language, while Rust grew fastest among the top languages and doubled in a year.
What agent-led coding means for business
Companies that ship software can move more review and routine implementation to agents while keeping engineers for approval and architecture. Agent drafts reached 110 million in September alone, up from a 17% share of pull requests in March to more than 40% by September. Activity rose sharply around June, when agents began working across days and running in the cloud. In September, 61% of Copilot users who selected their own model used more than one, so multi-model routing fits existing behavior.
Cost and infrastructure terms still require careful checking before wider rollout. GitHub added 4 million CPU cores, 160 PB of high-speed storage and 50 Tb/s of network capacity this year, and is moving to multi-region deployment on Azure with completion expected early next year. Existing Copilot Business and Copilot Enterprise customers can keep plans or switch, so licensing paths differ during transition. The news alone does not prove review quality in a specific codebase, despite the ReviewBench result for high-severity issues.
The marker to watch is whether the agent share of pull requests keeps rising after September and whether credit-based Enterprise adoption expands beyond sales-led deals. A second signal is completion of the Azure multi-region move early next year and its effect on resilience. If both hold, agent-initiated development will become the default operating pattern for commercial software teams.
