GitHub Universe 2026: Copilot app, Fusion model routing and a rebuilt core for agentic delivery

At its Universe 2026 event, GitHub set out a plan to become the most complete end-to-end system for delivering software, pairing a set of new Copilot products for agent-driven work with a rework of the infrastructure underneath the platform.

According to the GitHub Blog, the company sees the software development lifecycle as no longer a straight line from planning to deployment. A code change, a colleague's request or a production metric can set off work anywhere in the system. Developers set direction and remain accountable for what ships; agents implement and verify changes; and the results feed a learning engine intended to make the next cycle better.

An agentic delivery system with three goals

GitHub says the system is organized around three outcomes. Velocity means getting from idea to finished software as quickly and cheaply as possible. Intelligence means combining context from across the platform with a learning engine, so the system improves with use. Quality means completed work ships with security built in and bugs caught before release.

The Copilot app takes shape

The centerpiece is the GitHub Copilot app, designed specifically for agentic delivery. According to GitHub, it connects agents and tools across projects so developers can direct work without handling every handoff, and it covers more than code: writing specs, running design discussions and analyzing telemetry all happen in a single interface built around agents.

Copilot projects, available natively in the app starting at the event, let users give an agent a goal and run work across multiple repositories in managed parallel sessions, with one view for tracking progress and deciding what happens next.

A related feature, session continuity, keeps a single cloud agent session alive across the GitHub mobile app, VS Code, the CLI and the web: start something from a phone, steer it from an editor, check on it from github.com.

Fusion routes work across models

Copilot Fusion takes on model selection. Rather than making developers manage model choices themselves, Fusion evaluates what a task needs across reasoning, code generation, debugging and tool use, then decides how to execute it across one or more models. GitHub reports that on the Artificial Analysis Coding Agent Index, its Fusion Max configuration scored within two points of Opus 5.5 Max at half the cost and 40% less time per task.

Review and automation learn from results

Two further announcements target quality and repetitive work. Copilot code review Max, a deeper tier alongside the existing Lite and Balanced modes, posted the highest F1 score on ReviewBench for high-severity issues, according to GitHub. Copilot automations, which run AI-powered workflows inside a repository on manual, event or schedule triggers for tasks like issue triage, can now learn from their outcomes and propose improvements for a human to review and approve.

Copilot folded into enterprise plans

Pricing is changing as well. As of the announcement, GitHub Enterprise customers can use all Copilot features, including the app, code review and automations, without buying a separate subscription, paying instead for the AI credits they consume. The option is available through GitHub Sales immediately, with self-serve access promised soon; existing Copilot Business and Copilot Enterprise customers can keep their current plans or move over when ready. GitHub also stressed that the platform remains extensible, pointing to continued support for tools such as Cloudflare, Datadog and Figma.

The platform underneath is being rebuilt

GitHub argues that agents are changing both the volume and the shape of software development, and that adding capacity alone will not keep up. This year, the company says, it has added 4 million CPU cores, 160 PB of high-speed storage and 50 Tb/s of network capacity, and it is migrating to a multi-region deployment in Azure that it expects to finish early next year.

The company is also rearchitecting Git around three principles: simpler primitives, storage separated from compute, and minimal coordination between components. Similar work, it says, is underway across its AI gateway and agent harness.

Why it matters

The post was written by Jay Parikh, Microsoft's Executive Vice President of CoreAI, whose organization includes GitHub, a reminder of how central the platform is to Microsoft's agent strategy. His framing is stark: all software written to date will amount to less than 1% of the software written in the next five years. Whether or not that prediction holds, the announcements point to two concrete shifts. Bundling Copilot into Enterprise plans and metering by credits moves AI developer tooling from a seat-based add-on toward usage-based pricing. And the decision to redesign Git itself is a signal of scale: if agent-driven traffic is already straining the current architecture, machine-written code is no longer hypothetical on GitHub. For teams, the near-term questions are more mundane, what the credit-based model costs in practice, and whether code review Max and Fusion hold up outside the benchmarks GitHub cited.