Friday Work

Product Introduction

1. Definition: Friday Work is an open-source, MIT-licensed multi-agent coding harness and orchestration platform. Technically, it is a local desktop application (or Node.js CLI tool) that integrates with and coordinates multiple AI-powered coding assistants (agents) like Claude Code, Codex, and Cursor.
2. Core Value Proposition: It exists to solve the fragmentation of using multiple AI coding tools. Instead of agents running in isolated terminals, Friday makes them work as a cohesive team, enabling automatic task hand-off at usage limits, inter-agent communication, and collaborative task management under unified human review. Its primary keywords are multi-agent coding harness, AI agent orchestration, and open-source coding assistant.

Main Features

1. Intelligent Task Hand-off: When one AI agent (e.g., Claude Code) hits its usage limit mid-task, Friday automatically passes the work to another configured agent (e.g., Codex). It provides context via a structured brief containing the original task, recent code changes, file reads, command outputs, and test results, allowing the next agent to continue seamlessly.
2. Unified Task Board & Agent Messaging: Friday provides a central board where a single coding task can be broken down and assigned to different agents based on their strengths. Agents can send direct messages to each other (e.g., a heads-up about a file rename) within the platform, eliminating the need for manual copy-pasting between separate applications.
3. Git-Centric Review & Workflow: Each agent operates in its own isolated Git worktree. All proposed changes are consolidated into a single diff for human review. Reviewers can leave inline comments on specific code lines and stage changes hunk-by-hunk before committing, ensuring no code ships without explicit approval.
4. Multiple Operational Modes (Code, Work, Crew):
   • Code Mode: For parallel coding agents reviewed before commit.
   • Work Mode: For collaborative non-code work (plans, docs) between agents like Claude Code and Codex in a shared folder.
   • Crew Mode: Creates persistent AI "teammates" with names, roles, and memory. A lead agent decomposes a high-level goal, assigns subtasks, and manages the workflow.

Problems Solved

1. Pain Point: Context loss and manual overhead when switching between AI coding tools. Developers waste time re-explaining tasks and copying context between Claude Code, Cursor, and other terminals.
2. Pain Point: Work interruption due to individual agent usage limits. Hitting a daily limit on one tool halts progress entirely.
3. Pain Point: Lack of governance and review for AI-generated code. It's challenging to oversee and approve changes made by multiple, independent agents.
4. Target Audience: Full-stack and software developers who actively use multiple AI coding assistants; tech leads managing AI-assisted workflows; product teams using AI for both code and documentation.
5. Use Cases: Building a complex feature by having Claude Code design the architecture, Cursor implement UI components, and Codex write backend logic, with automatic hand-off. Maintaining progress on a refactor when Claude Code's limit is reached, passing it to Gemini CLI. Reviewing all AI-suggested changes from a sprint in one unified git diff before merging.

Unique Advantages

1. Strengths & Limitations (Pros & Cons):

   • Pros:
     • True Agent Interoperability: Unique focus on enabling different agents to communicate and hand off work, not just run in parallel.
     • Developer-Centric Control: Git-based review workflow gives final authority to the human developer, mitigating AI hallucination risks.
     • Cost & Privacy Efficient: Uses existing tool subscriptions; runs locally; no central API fees or data exfiltration.
     • Open Source & Extensible: MIT license allows inspection, customization, and support for new agents via the Agent Client Protocol (ACP).
   • Cons:
     • Setup Complexity: Requires local installation and configuration of each individual AI agent (Claude Code, Cursor, etc.).
     • No Cloud Sync: Work is tied to the local machine, which may hinder collaboration across different developer environments.
     • Dependent on Underlying Agents: Cannot surpass the inherent capabilities, limits, or costs of the configured tools.

2. Key Alternatives & Differentiation:

   • vs. Using Agents Individually: The manual alternative involves constant context switching and copy-pasting. Friday automates coordination, providing a unified interface and persistent context.
   • vs. Conductor: While both orchestrate AI agents, Friday is distinguished by its strong focus on git-native review workflows, inter-agent messaging, and a free, open-source, local-first model. Conductor may offer more cloud-centric or enterprise-focused features.
   • vs. T3 Code or Single-Agent IDEs: These are integrated environments for a single AI. Friday is a meta-tool that coordinates across multiple such environments, making them collaborative.

Frequently Asked Questions (FAQ)

1. Is Friday Work free and how does it make money? Yes, Friday Work is completely free and open-source under the MIT license. It does not currently charge users. The company may offer paid "Friday for teams" features in the future, but the core multi-agent harness remains free.
2. How does the hand-off between AI agents work without losing context? Friday does not stream the full conversation history. It creates a concise technical brief for the next agent, including the original task, the last action, files read, commands run (and their results), and code changes made. The receiving agent reviews this and re-runs relevant tests before proceeding.
3. Does Friday Work require new API keys or subscriptions? No. Friday operates the desktop applications (e.g., Claude Code app, Cursor IDE) you already have installed and are signed into. It leverages your existing plans and does not require separate API keys for its core orchestration function.
4. Can I use Friday Work if I only have one AI coding assistant (like Claude Code)? Yes. You still benefit from the unified task board, git worktree isolation, and structured review workflow. A task will pause when your single agent hits its limit and can resume once the limit resets, maintaining context within Friday.
5. What data does Friday Work collect and send? According to its documentation, the app sends minimal anonymous usage counts (e.g., version number) for update checks, which can be disabled in Settings. It explicitly states it never collects your code, prompts, or project files, as all data stays locally on your computer.