Managing a swarm of AI coding agents often feels like juggling chainsaws in the dark. That was the reality for a Checkly engineer running up to 20 concurrent sessions, including subagents. The solution wasn't another CLI wrapper or a chat interface, but a spatial map. The engineer built Agentic Crafting Table (ACT), an open-source desktop application that visualizes agent workflows on a zoomable canvas. Since adopting this tool, the developer reports merging up to twice as many pull requests per week compared to the previous six months.
The Problem With Terminal Multiplexers
Standard tools failed to scale with the workload. Terminal multiplexers worked for two agents but became unreadable at six, requiring constant text scanning to determine if an agent was waiting for input. Chat applications offered better visuals but disconnected context, scattering tickets, Slack threads, and RFCs across different windows. The core need was a unified layer that kept supporting evidence and agent status in one place, without adding complexity on top of existing tools like GitHub or Linear.
How ACT Works
ACT functions as an Electron app built with TypeScript and React, using React Flow for the map interface. Each project is a 'neighborhood,' and every taskβwhether an agent session, ticket, or PRβis an item on the map. The tool integrates directly with Claude Code and Codex sessions via an MCP server, allowing agents to read the map and post their own outputs. A 'Needs me' panel aggregates all items requiring human attention, such as agent questions or failed rollouts, eliminating the need for manual polling.
Development and Community Adoption
The tool was built using an iterative 'vibecoding' approach, where the engineer described problems to Claude and refined broken prototypes. While the author notes performance issues on large maps and unfinished concepts like the 'undermap' archive, colleagues at Checkly have already contributed significant features. These include Windows and Linux support, GitHub issue integration, and keyboard shortcuts. The project remains open source, encouraging users to build plugins for specific workflows, such as integrating Figma or Teams.
Key Takeaways
- ACT replaces terminal polling with a spatial map, doubling PR merge rates for the creator.
- The tool uses MCP servers to let agents interact with the map, posting their own results and status.
- Built on Electron and React, it keeps data local via SQLite and OS credential stores.
- Checkly colleagues contributed Windows support, GitHub integrations, and UI improvements.
The Bottom Line
While the UI still has rough edges, ACT proves that spatial visualization beats linear logs for managing agent swarms. Itβs a necessary hack for those running more than five agents at once.