Agents

GitHub Universe 2026: Copilot projects and code review Max, Factory's Droid 1.0, OpenCode taps Claude Code: the October 31 roundup

GitHub lets Copilot run agents across several repos and adds a multi-model Max review. Factory's Droid 1.0 has its agent write programs to chain tool calls.

By Audryk · October 31, 2026

• Coding agents
• Cloud & infrastructure
• #GitHub
• #GitHub Copilot
• #code review
• #agents
• #Factory
• #OpenCode
• #Claude Code

Two ideas dominate this week's agent news: one person directing many agents at once, and reviews strong enough to trust what they produce. GitHub pushed both at Universe on October 28. Factory answered on October 29 with a new version of its agent engine.

GitHub Copilot projects: one goal, several repos, several agents

In the GitHub Copilot app, a project gathers a goal and its context: repositories, design mockups, Slack threads or meeting notes. Copilot then runs "as many agents as the work needs" and brings back results such as pull requests for review. Projects are in public preview (GitHub Changelog).

The same release adds computer use in public preview on macOS and Windows: Copilot can read and click through desktop apps that have no API or MCP integration, after asking for approval. Organizations can switch it off through managed settings.

GitHub also announced session continuity (one cloud agent session followed from the app, VS Code, the CLI and the web) and Copilot automations that can propose improvements to themselves based on their results, for a human to approve (GitHub).

Code review Max and approvals that count

Copilot code review gets a third effort level, Max, next to Lite and Balanced, in public preview. It runs several independent review agents on models from different providers, then an aggregation agent reconciles their findings. GitHub says it takes longer and uses more AI credits (GitHub Changelog). GitHub says Max posted the highest F1 score on ReviewBench, its own open benchmark, for high-severity issues.

Copilot approvals are now generally available. When an admin turns them on (they are off by default), Copilot's approval counts toward a repository's required approvals and is dismissed if new commits are pushed. Repository admins can limit which file paths Copilot may approve. Both features need Copilot Pro, Pro+, Max, Business or Enterprise.

Copilot without a separate enterprise subscription

GitHub Enterprise customers can now use all Copilot features, including the app, code review and automations, without buying a separate Copilot subscription: they pay for the AI credits they use. It goes through GitHub Sales for now, with self-serve "coming soon" (GitHub).

GitHub also gave rare infrastructure figures: this year it has added 4 million CPU cores, 160 PB of high-speed storage and 50 Tb/s of network capacity, and it expects to finish migrating to a multi-region Azure deployment early next year. The context is in our Octoverse 2026 article: agents now start more than 40% of pull requests.

Factory Droid 1.0: the agent writes the program

Factory, which does not train its own models, sells the "harness": prompts, tools and the loop around the model. Droid 1.0, published on October 29, turns on programmatic tool calling by default. The model writes a JavaScript program that runs in a sandbox with no file, network or system access of its own and calls Droid's tools; raw results stay in the program and only what it prints goes back to the model. Factory says one script can make hundreds of tool calls in a single model request (Factory).

Factory's own Terminal-Bench 4 runs (October 23 to 28) put Droid 3.5 points ahead of Codex at 12% lower cost with GPT-6.1 Sol, and 3.0 to 7.6 points ahead of Cursor, OpenCode, Devin and Claude Code at up to 24% lower cost with Claude Opus 5.5 (Factory). These are vendor measurements with methodology choices of their own, such as cost counted only on tasks both harnesses passed. They have not been independently checked.

Tool updates

• OpenCode 2.2.3 (October 30) adds a built-in Claude Agent provider: install Claude Code, run claude auth login, then pick Claude Agent in /connect to use your Claude Code login and models. TUI plugins now require OpenTUI 0.8 (release notes).
• Claude Code v2.1.316 (October 31) adds CLAUDE_CODE_SUBAGENT_EFFORT to set the effort level of subagents and shared group budgets in the Claude apps gateway. Fast mode for Opus 4.8 is deprecated and will be removed on November 30, 2026; Anthropic points to Opus 5.5 instead (release notes).

What to keep in mind

• Copilot projects, computer use and code review Max are previews; only approvals are generally available.
• An approval that counts toward branch protection is a governance decision: it is off by default and can be limited to some paths.
• Factory's and GitHub's benchmark claims come from the vendors themselves.

See also: Copilot Fusion and Agent Merge, the other GitHub launches of the week.

Sources

• GitHub, "GitHub Universe 2026: Evolving every layer of the stack", October 28, 2026, checked October 31, 2026
• GitHub Changelog, Copilot app Universe releases, checked October 31, 2026
• GitHub Changelog, Copilot code review Max and approvals, checked October 31, 2026
• Factory, "Droid 1.0: Building a Frontier Harness", October 29, 2026, checked October 31, 2026
• Factory, Terminal-Bench 4 harness comparison, checked October 31, 2026
• OpenCode v2.2.3 release notes, checked October 31, 2026
• Claude Code v2.1.316 release notes, checked October 31, 2026

Spotted an error in this article? Report it · Our corrections policy