Summary
Max Drake, product engineer at tldraw, presents at the AI Engineer conference on the emerging paradigm of running AI agents directly on an infinite canvas. tldraw is both a widely-used free whiteboarding application and the underlying open SDK that powers canvas experiences at companies including Replit, and Drake argues that the spatial, multiplayer nature of a canvas makes it a uniquely productive environment for human-agent collaboration — one that mirrors how people naturally think and organize ideas.
The talk covers three progressively complex agent experiments built at tldraw. The first demonstrates spatial awareness: an agent navigates the canvas autonomously, adjusts its viewport, reads context, and modifies elements in response to natural language — analogous to how a coding agent searches a codebase. The second, called Fairies, introduces persistent multi-agent entities with visual identities (customizable hats, colors, a ‘leg slider’) that can collaborate with each other and with human users simultaneously on a shared multiplayer canvas. The third is a live end-of-talk demo where Drake asks an agent to locate a Notion document from a colleague’s email and build the described project directly in the tldraw desktop app.
Drake articulates a key insight throughout: LLMs were trained on text and produce text, but canvas work is inherently spatial and visual — and bridging that gap requires giving agents structured access to canvas state rather than raw pixels. The tldraw SDK handles this abstraction, enabling agent-canvas integration that would otherwise require building the entire canvas infrastructure from scratch.
📺 Source: AI Engineer · Published September 10, 2026
🏷️ Format: Keynote Launch
