← Back to context

Comment by worldsayshi

7 months ago

I think that "Figma/Excalidraw-like" canvas for <insert interesting problem domain here> is an underrated solution space. At least in terms of variation on the problem domain.

I agree, I especially find it useful when prototyping. I recently built a canvas node-based tool that connects LLM prompts and passes the model's response into the next node as context. It was fun to create and great for testing/comparing outputs from different models. https://littleworkflow.com/

(should add TLdraw which is specifically built for extensibility)

no, it is overrated. there are many many canvas workflow builders like this one. they're all beautiful, intuitive and easy to use. they all also dont get traction. go ahead, name one that everyone else here will know. its weird but theres a huge gulf here between what people think they want and what people really use outside the figma/excalidraw/tldraw/miro/canva space

  • we bet that its just because they pre-dated LLMs and don't really allow for granular control. the goal is to make it easier to develop automations/workflows, not harder. we think the future of development is 90% transformations/routing/fallbacks (boring) and the only part that actually impacts the end result is the call to the LLM (fun), the tools you give it, and a memory component. if we abstract away all the boring stuff, and only leave the stuff that impacts the results then it makes iterating and building with AI much faster and easier.

We believe a visual solution is the best way for developers — and LLMs — to build applications in the future.