Friday Introduces Multiplayer Workspace for AI Coding Agents

New desktop app lets teams run Claude, Codex, and Cursor side by side on shared context.

GeoDetect PipelineOctober 7, 2026
Share

Overview

GitHub on Friday introduced Friday, a new desktop application designed to facilitate collaborative coding sessions between human developers and multiple AI agents. The platform allows teams to run tools such as Claude Code, Codex, and Cursor side by side within a single, shared context. This approach aims to resolve the isolation that typically characterizes individual AI-assisted development workflows.

The central premise of Friday is that coding agents should not operate in silos. Instead, teammates can watch, steer, and take over control of different agents simultaneously. A key architectural decision is that all code execution and agent processing remain local to the user’s machine. No source code is uploaded to external servers for processing, addressing significant privacy and intellectual property concerns for enterprise users.

How Friday Works

Friday functions as a desktop app available for Windows and Mac, allowing users to link their existing coding environments. When a session starts, the application detects locally installed agents like Claude Code, Codex, and Cursor. Teammates can join the session via a shareable link, bringing their own agents into the same workspace.

The interface provides a live feed of activity, showing which teammate is driving which agent at any given moment. In a typical scenario, one developer might direct Claude Code to investigate a bug, while another uses Codex to write a failing test for the same issue. Both agents operate on the same codebase context, ensuring that their actions are coordinated rather than conflicting.

Control of each agent is strictly managed to prevent overlap. Only one person can drive a specific agent at a time, but different agents can be controlled by different team members simultaneously. This structure mimics the traditional pair programming model, scaled up to include multiple AI assistants and several human participants.

"Your team and your coding agents, in one live session. Run Claude Code, Claude Code, Codex and Cursor side by side on the same context."

Key Features and Workflow

The platform introduces several features designed to streamline collaborative AI coding:

  • Shared Context: Every participant sees the same code state. If one agent modifies a file, others see the change immediately.
  • Manual Mode Approvals: Agents can be set to request approval before running commands. These requests are visible to the entire team, and the current driver must approve them.
  • Session Replay: Every action, question, and tool response is recorded. Users can rewind the session to any point to review decisions or retrace steps.
  • Cross-Tool Integration: GitHub, Slack, and Linear connections are set up once per session, allowing all connected agents to access these services without individual reconfiguration.

The workflow supports three levels of agent autonomy: Plan, Manual, and Auto. In Plan mode, the agent reads and proposes changes but touches nothing. Manual mode allows file edits but requires approval for every command execution. Auto mode grants full freedom for tasks the team trusts implicitly.

Pricing and Availability

Friday is available as a free download for individual users. The free plan includes the core collaborative features, allowing users to bring their own coding agents without metered restrictions. The service emphasizes that it does not store code, relying instead on local execution.

For teams looking to standardize practices, Friday offers a free browser-based generator to create configuration files such as CLAUDE.md, AGENTS.md, and Cursor rules. These files define the behavior and constraints of the agents, ensuring consistency across the team without sending project details to external servers.

The company positions Friday as a solution to the friction caused by switching between tools or handing off work. By keeping everyone in a single live session, the platform aims to preserve context and provide a clear audit trail of who asked for what, and what each agent did.

Share

Sources

Tools mentioned