Spine Swarm Uses Visual Canvas Instead of Code to Coordinate AI Financial Agents

Spine Swarm, a YC S23 startup, has built a multi-agent AI system where agents collaborate on an infinite visual canvas rather than through traditional code execution or API calls. Each agent claims a spatial region on the canvas and writes structured artifacts — such as spreadsheet tabs, charts, or slides — that other agents can discover and build upon. The architecture replaces conventional message queues and function calls with spatial layout and artifact persistence, making it better suited to financial modeling workflows. An append-only event log tracks every artifact creation, mutation, and cross-reference, enabling replay of the canvas state at any point in time. A continuous validator agent monitors the canvas for inconsistencies and spawns repair tasks, which is central to maintaining accuracy in complex outputs like market-sizing models and earnings trend reports.