Written by Henk Boelman on

GitHub Universe 2026: from coding assistant to software delivery system

At GitHub Universe 2026, Copilot grew from a coding assistant into a full software delivery system, with Copilot Fusion, Copilot projects, shared automations and Copilot included in GitHub Enterprise. We bundled the Day 1 announcements into one overview, with a community view at the end.

8 min read

GitHub Universe kicked off yesterday in San Francisco, and the message was clear: GitHub no longer sees Copilot as an assistant in your editor, but as the engine of a complete software delivery system run by developers and their agents. In the keynote post, Jay Parikh, Executive Vice President of CoreAI at Microsoft, put it simply: the goal is completed work, not more code.

Keynote GitHub Universe Day 1

There was a lot to take in, so we packed the Day 1 announcements into one overview, with a community view at the end.

The big picture: agentic software delivery

GitHub argues that the SDLC (software development life cycle) is no longer a straight line from planning to coding to testing to deployment. Work can start anywhere: a code change, a request from a colleague or a production metric. Developers set direction and stay accountable, while agents implement and verify changes.

GitHub frames its new system around three goals: velocity (get from idea to finished software fast and cost-effectively), intelligence (use context from across the platform and learn from every cycle) and quality (secure by default, bugs caught before they ship). Almost every announcement maps to one of these.

Working with agents

  • GitHub Copilot app. The agent-first desktop app GitHub introduced earlier this year got a major update. It now aims to cover the whole path from issue to merge, and beyond code you can write specs, run design discussions and analyze telemetry in the same place, using context from connected tools like Slack and Figma.
  • Copilot projects. Give Copilot a goal and it runs agent work across multiple repositories in managed, parallel sessions, with one place to track progress and make decisions. Projects are in public preview in the Copilot app.
  • Session continuity. Start a cloud agent session on your phone, steer it from VS Code (Visual Studio Code) at your desk and check in from github.com later. The same session follows you across the app, VS Code, the CLI (command line interface) and the web.
  • Computer use. In public preview on macOS and Windows, Copilot can work through tasks in your desktop apps, even ones without an API (application programming interface). It asks for your approval before taking control of an app, and organizations can switch the feature off.
  • Image generation. Copilot can now generate and edit images from a chat prompt and save them straight into your project, in public preview. Organization owners can switch it off, and an optional "Made with AI" watermark is available.

Picking the right model: Copilot Fusion

Copilot Fusion is now generally available. It combines the multi-model orchestration from Project HydraFusion, a research preview GitHub released in September, with Copilot's automatic model selection. Instead of picking a model yourself, you pick an effort level (Standard, High or Max), and Fusion decides per prompt which model or combination of models to use. According to GitHub, Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of Claude Code with Opus 5.5 Max, at about half the cost and tokens per task and about 40% less time.

In an opt-in experimental preview, Fusion can now also route between on-device and cloud models in the Copilot CLI, the Copilot app and VS Code. On supported Windows devices it can download and set up Microsoft's MAI Code 1.1 Flash model locally, so some requests run on your own hardware.

Context that sticks: GitHub intelligence

GitHub intelligence is a new context and learning layer built on the data GitHub already holds about your work. It gives agents a memory of your preferences and repository conventions, keeps that guidance up to date over time and lets agents search code and history beyond the files in front of them. The goal is that agents build on what your team already knows instead of starting every session from scratch.

Automating the repetitive work

  • Copilot automations. AI-powered workflows in your repository that run on demand, on an event or on a schedule, for tasks like reproducing bugs, issue triage and code review. Two additions: automations can now learn from their results and propose improvements for you to approve, and shared automations let a whole team build and maintain them together, billed to the organization instead of one person.
  • GitHub Agentic Workflows, generally available. This open source project from GitHub Next lets you describe automations in plain Markdown, which are then compiled into standard GitHub Actions YAML files. Think CI (continuous integration) failure analysis, documentation updates or test writing. You can use Copilot or bring your own coding agent.

Quality and security

  • Copilot code review Max. More code means more pull requests (PRs) to review. Next to the faster Lite and Balanced modes, the new Max mode takes a deeper look when a change needs it. GitHub says it posted the highest F1 score, a measure that balances catching real issues against false alarms, on ReviewBench for high-severity issues.
  • AI Scan for repositories. In public preview, an agent scans your entire repository for vulnerabilities, not just new changes. It covers languages that CodeQL, GitHub's rule-based code analysis engine, doesn't support, and findings appear next to existing CodeQL alerts.
  • Security extension for the Copilot app. Browse your code scanning alerts by severity inside the app and ask Copilot to explain or fix them. Fixes stay in your local copy until you decide to commit.

Pricing: Copilot included in GitHub Enterprise

GitHub Enterprise customers can now get all Copilot features included, including the app, code review and automations, for every licensed user, without a separate Copilot subscription. Instead of managing seats, you manage one AI usage budget: usage is billed from the first request, with no monthly credits included. Code completion keeps working when the budget runs out, while features like Copilot Chat and code review need budget. For now this is available for GitHub Enterprise Cloud through a GitHub or Microsoft account team, with self-serve coming in stages. Copilot Business and Copilot Enterprise remain available.

The foundation underneath

Agents are changing the volume of development. GitHub recorded 7.38 billion commits in September 2026, more than five times as many as a year earlier. Its Octoverse 2026 report, released the same day, adds that more than 40% of pull requests in September started with a first draft written by an AI agent, up from 17% in March. To keep up, GitHub has added 4 million CPU cores, 160 PB (petabytes) of high-speed storage and 50 Tb/s (terabits per second) of network capacity this year. It's also moving to a multi-region deployment on Azure, expected to finish early next year, and rebuilding its Git architecture to separate storage from compute.

Our take as a community

Put next to Entire's launch on Monday, a pattern becomes visible. Both companies are betting on the same three ideas: review is the new bottleneck, agents need lasting context about how a team works, and the infrastructure underneath has to be rebuilt for agent-scale traffic. Their approaches differ, but the questions they answer are the same.

For builders in our community, a few things stand out:

  • You're becoming the orchestrator. Copilot projects and session continuity assume you direct several agents at once. How you plan, split and review that work is quickly becoming a core skill.
  • Context is a team asset. With features like GitHub intelligence, repository conventions and custom instructions directly shape agent output. Writing them down well is worth the effort.
  • Check the numbers yourself. The benchmark results come from GitHub's own announcements. Try the tools on your own code and judge the results there.
  • Watch your credits. Moving to pay-per-use AI credits makes costs flexible, but also less predictable. Keep an eye on usage while you experiment.

What's next

Universe continues today with Day 2. Have you tried the Copilot app, Fusion or Agentic Workflows yet? Share what you learned in the community channels or at your next chapter meetup or AgentCon stop. Hands-on sessions from community members are exactly what we love to put on stage.

Sources

Written by Henk Boelman

Keep reading

Global AI Community blog banner: “First glance at Entire.” Many AI agents feeding into one human review.

First glance at Entire: rethinking where code lives in the agent era

Entire wants to rethink where agent-written code lives. Here’s our first glance at the new Git forge from former GitHub CEO Thomas Dohmke.

Read the post →

Say hello to n8n, our newest partner!

n8n, the fast-growing workflow automation platform from Berlin, is now a partner of the Global AI Community.

Read the post →
See all blog posts