AI Agent Task Manager

What It Does

AI Agent Task Manager is a free task board designed specifically for developers running multiple AI coding agents simultaneously. Instead of juggling terminals and losing track of which agent is stuck or finished, the board provides a single visual map of every parallel coding session — Claude Code, Codex, Cursor — each in its own workstream column.

Who It's For & The Problem It Solves

Built for solo developers and small teams running 5+ concurrent AI agent sessions. The core problem: when you spin up multiple coding agents across terminals, you quickly lose track of which one needs your input, which is working autonomously, and which finished an hour ago. The typical workflow — bouncing between tmux panes, Notion docs, or scattered terminal tabs — wastes time and mental overhead. AI Agent Task Manager replaces that chaos with a centralized board where each workstream has its own queue, clear status indicators, and browser-level push notifications when an agent genuinely needs you.

Key Features & Capabilities

• Workstream-per-Agent Architecture: Each column is one agent session (up to 24 parallel streams). Queue tasks with full instructions; the agent pulls the top card, works autonomously, marks it done, and pulls the next.
• Smart "Needs You" Signaling: When an agent hits a blocker requiring a decision, the column turns amber and fires a browser notification — you stop hunting across terminals for the stuck session.
• Bring Your Own Agent: Works with Claude Code, Codex, Cursor, or any framework that can run curl. Connects via a tiny REST protocol (pull/attention/resume/complete) behind a personal API key.
• Queue Management: Each workstream maintains its own ordered queue. Use "Make next" to bump urgent cards to the front mid-run.
• No Lock-In: The agent-side protocol is four REST endpoints (GET next task, POST attention/resume/complete). Wire your own scripts or use the copy-paste bootstrap prompt.
• No Signup to Try: Open the board and start queueing tasks immediately; sign in only when you want to connect agents and sync across devices.

Typical Workflows

• Parallel Feature Work: Run three workstreams (front-end, API, tests) with three Claude Code sessions. When the API agent blocks on a migration decision, that column lights up and pings you; the other two keep running.
• Overnight Backlog Grind: Queue 15 small chores on one workstream, paste the bootstrap prompt before logging off, wake up to most done with one-line summaries.
• Big Refactor + New Features: One agent handles a slow, careful refactor on one branch; another ships a fast feature on a separate branch — both running in parallel without interference.

Integration & Use

Connect agents via a copy-paste prompt and personal API key. The board exposes a REST protocol agents poll autonomously. No external integrations required — it's a standalone web app designed to orchestrate your existing AI coding toolchain.

What Differentiates It

Unlike Notion (external memory only, no API for agents), Trello/Linear (built for human assignees, not autonomous polling), or terminal multiplexers (sees panes, not progress), AI Agent Task Manager is purpose-built for the workflow of running multiple autonomous coding agents at once. It's the missing semantic layer — status, queue, and "needs you" signals — for developers who've outgrown one-agent-at-a-time.

Prompting Patterns

The board encourages best-practice task instructions: lead with the outcome (not steps), name file paths explicitly, define "done" criteria, list constraints, link prior art, and keep one task per card. Ready-to-paste prompts are provided for mechanical refactors, cleanup passes, feature work, and open-ended investigations.