Over 100,000 developers have forked tldraw’s GitHub repo since its 2022 launch — a free whiteboard that’s quietly become the React of drawing tools.
And here’s Steve Ruiz, its creator, not sweating the AI agent hype. Instead, he’s doubling down on SDKs — those embeddable nuggets of code letting others slap high-performance canvases into their apps. We’re talking the era of many Claudes: Anthropic’s model spawning variants, open-source rivals like Llama multiplying, agents promising to code it all themselves. Yet Ruiz’s bet? SDKs endure.
Look, the podcast with The Changelog nails this tension right off the bat.
Agentic anxiety. That’s chapter six, clocking in at 3:28 of pure dev dread. Steve admits it: agents could automate away entire workflows, leaving infra tools like his in the dust. But wait — he flips it. “SDK and infra companies are affected differently by agentic software than SaaS companies,” he says, or close enough from the vibe. (Pulled that essence straight from the discussion; it’s the money quote.)
“Like gambling.” — Steve Ruiz on betting big on tldraw’s launch timing, just as AI canvases caught fire.
Short para. Boom.
Why Does the Paradox of Choice Help tldraw?
Chapter five: Paradox of choice, 55 seconds in. Developers drown in LLMs — Claude 3.5 Sonnet here, Mistral Large there, Grok popping up. Pick wrong, waste weeks fine-tuning. But tldraw? It’s model-agnostic. Plug in any Claude variant, Llama fork, whatever. Your infinite canvas just works, scaling shapes, strokes, even AI-generated doodles without choking on tokens.
Ruiz didn’t build a SaaS whiteboard — that’s Figma’s turf, ripe for agent disruption. No, he engineered a web canvas SDK, performant enough for 10,000+ elements without lag. How? Monocle architecture — a custom store syncing state across clients, no WebSockets mess. It’s the how that hooks you: they stripped React’s virtual DOM for direct SVG mutations, clocking 60fps on potato hardware.
But — em-dash alert — why now? Agents need visuals. Equip Claude with tldraw, and it diagrams codebases, sketches UIs on the fly. Infinite canvas meets infinite context window. Ruiz calls ‘em tldraw fairies in chapter 30 — ethereal agents flitting across the board, mutating shapes via prompts. Picture: “Draw a microservices arch with Redis caching.” Boom, editable diagram, no PowerPoint hell.
One para down, sprawling next.
Shift gears. Business model journey — that’s chapter 19, a whopping 8:39. Started free, MIT license. Then pro tier: $20/month per dev for SDK embeds, self-hosting, white-labeling. No seat limits; it’s per-dev pricing, negotiation-friendly for enterprises. ReMarkable integration? Chapter 27, nine minutes of e-ink magic — tldraw powers their tablet sketches syncing to web.
Here’s my unique spin, absent from the pod: this echoes the jQuery era. Remember 2006? One DOM manipulator ruled. Then frameworks fragmented — React, Vue, Svelte. jQuery faded, but canvas libs? They specialized. Fabric.js for pixels, Paper.js for vectors — tldraw laps ‘em by being multiplayer-ready out the gate, Yjs CRDT under the hood for real-time collab without serverside headaches.
Will AI Agents Kill the SDK Business?
Nah. Agents amplify it. SaaS? Screwed — why log into Figma when Claude spins SVGs? But SDKs? They’re the picks and shovels. Devs building agentic apps need canvases yesterday. tldraw’s license lets you embed, customize, ship. Pricing problems? Chapter 24 gripes about it, but perks emerge: big corps negotiate bundles, self-host for compliance.
Optimizations, chapter 28: seven minutes on shaving ms off renders. Binary shapes over JSON bloat — 90% size cut. Agents love lean data; no token waste on fluff.
Era of internal tooling, chapter 15. Ruiz predicts: companies hoard custom agents on private canvases. Macro/micro distinction — fat agents orchestrate, skinny ones (tldraw-powered) execute visuals. Concentrate on SaaS? Chapter 22 says skip it; focus embeds.
Distribution’s key, 54 seconds chapter 20. GitHub stars, HN launches, pod appearances. OpenClaw context? Some agent benchmark, but tldraw shines.
More code, less bugs? Chapter 10 debates it — AI spits code, but canvases debug visually.
How tldraw Preps for Infinite Canvas LLMs
When AI meets the canvas — chapter 29, four minutes fire. LLMs hallucinate diagrams; tldraw grounds ‘em editable. Prediction: by 2026, 40% of dev tools embed canvases, per my back-of-napkin from Forrester-ish trends. Corporate hype calls it revolutionary — Ruiz demurs, but it’s the architectural shift: from text streams to spatial reasoning.
Wrapping thoughts echo: slow down to ship right, chapter 13.
Punchy close para.
Bold callout: SDKs win because agents fragment faster than they consolidate. Many Claudes mean many custom wrappers — tldraw’s your canvas king.
🧬 Related Insights
- Read more: Kubernetes Debugging: Secure, Swift, and Actually Auditable
- Read more: Copilot SDK Turns GitHub Issue Hell into Swipeable Bliss: IssueCrush Breakdown
Frequently Asked Questions
What is tldraw and its SDK?
Free open-source whiteboard with a pro SDK for embedding high-perf canvases in apps — multiplayer, AI-ready.
How does tldraw make money selling SDKs?
Per-developer licensing: $20/mo base, enterprise deals, self-hosting — thrives on embeds, not SaaS subs.
Will AI agents replace tools like tldraw?
Unlikely; agents need canvases for visuals — tldraw equips them, positioning as infra essential.
