GitHub launches Copilot Fusion and Projects to automate software workflows

From code completion to coordination

GitHub is expanding its role from a repository for source code into a platform that manages the entire lifecycle of software development. During GitHub Universe 2026, the company demonstrated how its new tools can take a business idea and turn it into a functional application. The goal is to move beyond simple code generation toward a system that understands business context and coordinates complex tasks.

The demonstration showed a fictional coffee loyalty app being built using the new GitHub Copilot app. The assistant was able to break down tasks, pull relevant designs from Slack and Teams, and interact with a web browser. It even spun up parallel agents to generate promotional materials like product videos. This shift suggests that the real work of development is moving away from manual typing and toward high-level direction.

Microsoft EVP Jay Parikh, who heads GitHub, stated that the vision is to join the journey from intent to completed software. He noted that agents will do more of the execution while developers focus on directing the results. This strategy aims to improve delivery speed and quality by automating the most tedious parts of the development process.

Optimizing models with Copilot Fusion

One of the biggest challenges in AI development is deciding which specific model is best for a given task. Using a high-end model for every job is expensive, while using a smaller model for a complex problem can waste time. Copilot Fusion solves this by allowing developers to estimate the effort required for a task so the system can select the most cost-efficient model.

The system uses several different execution patterns to complete work. It can use a single model for simple tasks, a cascade to escalate drafts to stronger models, or a critique process. In the critique process, one model reviews the work of another to find errors before the final revision is made. This prevents a single model from simply correcting its own mistakes without outside oversight.

According to iTWire, GitHub reported that Fusion Max scored within two points of Claude Code with Opus 5.5 Max on the Artificial Analysis Coding Agent Index. It achieved this while using roughly half the average API cost per task and taking about 40% less time. To encourage adoption, GitHub is offering a 30% discount on Fusion's model costs through the end of November.

Managing scale with Copilot Projects

While Fusion handles model selection, Copilot Projects addresses the problem of coordinating multiple agents across different repositories. Large enterprise projects often require many different components to work together simultaneously. Copilot Projects allows users to set several agents on a job while keeping all relevant documents, requirements, and repositories in one place.

A key feature of Copilot Projects is its ability to run independently in the cloud. This means a developer can start a complex task and close their laptop while the agents continue to work on the software. This is designed to make large-scale undertakings, like building a coffee-ordering feature across multiple systems, much more manageable for small teams.

GitHub COO Kyle Daigle noted that while generating code is becoming easier, getting software into production remains a difficult enterprise problem. The company is investing in the parts of the delivery pipeline that require trust and governance. By providing a common place to manage these complex workflows, GitHub aims to help developers focus on what should be built rather than how to organize the work.

Enterprise adoption and growth

GitHub is making these tools more accessible to large organizations by changing its subscription model. All new GitHub Enterprise subscriptions will now include Copilot access without requiring a separate second subscription. Existing customers are encouraged to speak with account managers to see how these changes affect their specific pricing plans.

The platform's growth is already showing a significant shift toward agent-based development. The 2026 Octoverse report shows that the user base has grown to more than 225 million developers. It also highlights a massive uptick in agent usage, with developers starting over 40% of pull requests with an agent in September 2026, up from 17% in March.

These updates signal that GitHub is positioning itself as the primary home for AI-driven software development. The company is not just providing a place to store code but is actively building the infrastructure to manage agentic AI. The goal is to produce quality, secure, and cost-efficient code while helping developers in every country turn their ideas into finished products.