CEO at @magicpathai ๐ŸŽจโœจ Previously, @AnthropicAI, @brexHQ. @Uber, @Facebook. Creator of Claude Engineer, DesignerGPT, Sequential thinking MCP and more

San Francisco
Joined September 2009
Introducing MagicPath, an infinite canvas to create, refine, and explore with AI. Create beautiful components and functional apps, while providing production ready code. Available today, free, for everyone. The Cursor moment for design is here.
Just shared the repo for this
12
Made some quality of life changes, and you can see the progress as the writing happens. Here Kimi one-shotting 30k characters for one chapter only lol
5
Open sourcing Kimi-writer. github.com/Doriandarko/kimi-โ€ฆ
Kimi-k2-thinking is incredible. So I built an agent to test it out, Kimi-writer. It can generate a full novel from one prompt, running up to 300 tool requests per session. Here it is creating an entire book, a collection of 15 short sci-fi stories.
A new way of designing thatโ€™s based more on raw feelings.
How I make interactions feel seamless in @MagicPathAI Start with a strong base Add image refs, use the @MagicPathAI extension, and prompt in detail. Refine the design Add extra prompts or use the โ€œEditโ€ tool to fine-tune specific sections. Prompt each keyframe/variant The more context you give (screens, wireframes, frame-by-frame descriptions/images), the clearer your intent becomes. Refine interactions Use simple prompts like "make it seamless, smoother, less laggy" and see the magic happen.
1
35
Here's the link to the full book: we.tl/t-eLDEGapMaJ
5
35
Kimi-k2-thinking is incredible. So I built an agent to test it out, Kimi-writer. It can generate a full novel from one prompt, running up to 300 tool requests per session. Here it is creating an entire book, a collection of 15 short sci-fi stories.
Build it and they will come
Vibe design is here. MagicPath just got Chrome Extension to grab live UI. Now anyone can capture any component, import into MagicPath, edit with AI or visually. 6 wild examples + how to try:
8
7
1
206
We really don't sleep at MagicPath lol
We shipped so many things at MagicPath over the past month or so that if you blinked you may have missed: 1/6 Editor V2: Bring Figma-level editing to your vibe-designed ideas (and yes, that's layers on the left, currently in beta ๐Ÿ‘€)
2
2
95
Soon we will launch our new model at @MagicPathAI. It will be the best model in the world for interface design. Design is, and always will be, the moat. Here are some one-shot results from an early checkpoint.
This is still in beta, and our recommendation is don't import really big parts or full websites, but do it on a component level (sidebar/different sections, etc.). We will improve it as we go. We are already working on some updates. We also made it really easy to access it from the app as well.
4
1
14
0
It's truly a magical experience, especially when you get to just pop a working element into your canvas. Our beta testers said it's one of the most powerful AI flows they have ever used. Some people are even using it as a way to clean up and actually properly design some of their already vibecoded apps. It's so much easier to visualize things this way.
1
12
What you get, of course, is an interactive React component and code you can export. You can also create a reusable component library directly from the captured element.
1
8
0
Inspiration also lives on the web. You can import anything you like and change/adapt it to your use case. And it works with anything, and I mean ANYTHING. You can now remix the web.
1
12
0
A big problem today is that AI can't really understand your brand or product. We've done a lot of work at MagicPath to solve this. First themes and component libraries, but your real design lives on the web, so now you can edit, prototype, and build with it. Here I am importing some components from @tryramp landing page.
1
14
0
Your real product lives in the browser. Introducing MagicPath web capture: Import any functional web element into a canvas, edit it visually or with AI, and even request changes right as you capture. You can now finally build with your real product. Capture. Edit. Ship.
The best part about building an AI design platform is that you get to build the thing in the thing. We build MagicPath in MagicPath, so what takes weeks elsewhere takes us hours. And it's a strong self-reinforcing loop.
This is experimental, but I think it shows a possible future: strong fine-tuning of local models on specialized datasets. The repo also includes: - Utilities for converting models to MLX - Inference tools - GRPO works with ANY model (not just nanochat!)
1
11