Developers no longer write every line themselves; they now direct AI assistants and coordinate fleets of agents, the company says. Despite that shift, the yardstick by which work is judged remains unchanged: completed software — a feature, a bug fix, a security fix. Shipped work, not lines of code, is the deliverable.

The software delivery system GitHub is building

GitHub's stated goal is to become the most comprehensive software delivery system in the world, a system built on the view that the SDLC is no longer a linear sequence of plan, code, test, and deploy. A code change, a colleague's request, or a production metric can originate work at any point in that chain, pushing developers and agents to move between stages as tasks demand. In this model, developers set direction and remain accountable for what ships, agents implement and verify, and every cycle feeds a continuous learning loop. GitHub frames the system around three properties: velocity, getting from idea to completed software quickly and at low cost; intelligence, combining platform-wide context with a continuous learning engine; and quality, making shipped work secure by default and catching bugs early. The underlying primitives are being made available as well, so teams can build bespoke solutions while keeping control of the work, tools, and cost.

Agent orchestration in the Copilot app

The GitHub Copilot app, designed from the ground up for agentic software delivery, links agents and tools across projects so developers don't have to manage every handoff. Its remit goes beyond generating code: writing specs, presenting design discussions, and analyzing telemetry all happen in one agent-first surface.

Copilot Fusion addresses model selection by evaluating what a task needs across reasoning, code generation, debugging, and tool use, then deciding how to execute across one or more models — the goal being the right result at the right price without manual per-model decisions. On the Artificial Analysis Coding Agent Index, Fusion Max scored within two points of Opus 5.5 Max at half the cost and 40% less time per task.

Copilot projects, available natively in the app, give Copilot a goal and run agent work across multiple repositories in fully managed, parallel sessions, with a single place to track progress and make decisions. Session continuity lets a single cloud agent session follow the developer across the GitHub app, VS Code, the CLI, and the web; the agent keeps running in the cloud while the developer steers from whatever surface is at hand. GitHub describes the multi-model harness enabling this as powerful, fast, and cost-efficient.

Automations and code review

Copilot automations create AI-powered workflows inside a repository that run in the cloud, triggered manually, by an event, or on a schedule — useful for coordinating agents on tasks such as issue triage and code review. Automations can now learn from their outcomes and propose improvements for a human to review and approve, an application of the continuous learning idea to real runs.

Review capacity has become the new bottleneck, since rising request volume follows closely on rising code volume. Copilot code review surfaces actionable bugs, and a new tier, Copilot code review Max, posted the highest F1 score on ReviewBench for high-severity issues. Lite and Balanced remain the fast everyday options, with Max as the deeper pass for changes that warrant it.

Pricing and platform openness

GitHub Enterprise customers can now use all Copilot features — including the Copilot app, code review, and automations — without a separate subscription, paying only for the AI credits they consume. The arrangement is available through GitHub Sales today, with self-serve coming soon, and existing Copilot Business and Enterprise customers may keep their current plans or move over when ready. GitHub also stresses that it remains open and extensible, with integrations such as Cloudflare, Datadog, and Figma continuing to work as before.

Rebuilding the foundation

The platform underneath all of this is being rearchitected, not merely expanded: agents are changing both the volume and the shape of development, and added capacity alone won't cover it. Capacity added so far this year:

  • 4 million CPU cores
  • 160 PB of high-speed storage
  • 50 Tb/s network capacity

GitHub is also migrating to a multi-region deployment in Azure while serving its existing global traffic load, with completion expected early next year. Beyond room to grow, the systems that carry the work are being rebuilt. Git, scaled for years against growing activity, is being reworked around three principles for the agentic era's higher throughput and reliability ceiling: simplify the primitives, decouple storage from compute, and minimize coordination. Similar work spans the AI gateway and the harness.

The opportunity isn't more code — it's problems that couldn't be solved before, GitHub argues, claiming everything written to date may amount to under 1% of the software to come in the next five years.