AI Engineer World's Fair 2026 — unofficial import demo

What if chat is the wrong interface for managing agents? What if we're holding ourselves back by squeezing our thoughts and the way we work to into a one-dimensional, single-threaded interface? At a high level, this talk aims to present the work we've done at tldraw to build a spatial harness, or a way to allow agents to work on a canvas and collaborate with users and each other natively. This work represents important steps towards building better agent + canvas experiences, a product category we've seen explode in the recent months (Paper, Replit Agent 4, Google Stitch, etc). It's also not something I've really seen talked about elsewhere. See: - Multi-agent collaboration on the canvas (fairies.tldraw.com) - We've also recently brought code mode (https://blog.cloudflare.com/code-mode-mcp/) to the tldraw desktop app and MCP app.