AgentCanvas: visual multi-agent task canvas

A visual battle map for developers directing multiple coding agents

Problem

Many developers no longer write code themselves but instead fire off tasks to multiple AI agents, wait for results, and reschedule. Tracking what each agent is doing, which task is stuck, and switching windows is chaotic—some have even built their own split-screen tool just to cope.

Solution

A spatial canvas where each agent task is a card showing status, elapsed time, and output summary. Users can drag to set dependencies, dispatch tasks to different agents with one click, and auto-collect results. All outputs appear on a single timeline for easy review and follow-up instructions.

Users

Individual developers and small teams who heavily use agentic coding tools like Claude Code and Codex

MVP

Local canvas app supporting manual task cards linked to terminal/browser agent sessions; manual or simple polling status updates; dependency lines and completion alerts; export task logs.

Revenue

Free base version plus subscription for multi-project spaces, team collaboration, and auto status sync; or a one-time desktop license.

Why now

AI coding is shifting from writing code to directing agents, making parallel task management the new normal. Existing IDEs and terminals weren't designed for multi-agent orchestration. An indie developer can ship a lightweight tool and grow through community sharing.