I replied to 50+ Claude Design posts this week. Here's what I learned before launching Forge.
I'm the CEO of UXPin. We launched Forge on Product Hunt today.
But before I talk about that, I want to share what the last week taught me about what people actually want from AI design tools.
Claude Design launched 2 weeks ago. My LinkedIn feed exploded. I spent the week reading every take, replying to dozens of posts, and testing Claude Design myself.
Five patterns showed up over and over.
1. Token burn is the #1 frustration
Multiple users maxed out weekly limits in 2-6 hours. One burned 25% on 7 slides. The community built a workaround strategy: Opus for first prompt, Sonnet for edits, Haiku for tweaks. When your users need a cost mitigation strategy on day one, that's a signal.
2. "Design system integration" disappointed people
Claude Design reads your codebase and extracts visual patterns. It doesn't use your actual components. Designers reported wrong fonts, incorrect colours, and spacing drift. One spent more time fixing the AI's output than building from scratch.
3. Manual edits cost credits
Every adjustment routes through the AI model. Spacing changes, colour tweaks, layout shifts - all burn tokens. One designer said: "If I could just drag this element manually, it would take a split second. Why burn tokens for something so trivial?"
4. The output is still a handoff
Claude Design exports to Canva, PDF, PPTX, or Claude Code. Claude Code then interprets the design and generates new code. Multiple developers said they still had to rebuild the design system in code afterwards.
5. Nobody thinks it replaces designers
This was the most consistent take. Strong first draft tool. Democratises visuals for non-designers. But taste, judgment, edge cases, error states - still human work.
These five gaps are exactly what we built Forge to solve:
→ Your component library syncs from Git. The AI is constrained to it. No drift.
→ Manual edits use real design tools, not AI credits. No token burn on refinement.
→ The export is production-ready JSX referencing your actual imports. No rebuild.
→ Built-in libraries (MUI, shadcn, Ant Design, Bootstrap) if you don't have a custom system yet.
→ AI handles the scaffold. Designers handle the craft. Same canvas.
We've been building toward this for over a year. The timing of Claude Design's launch was coincidental - but it made the problem we solve more visible than any marketing we could have done.
Forge is live on Product Hunt today:
https://www.producthunt.com/posts/uxpin-forge
If you tried Claude Design this week and felt the gaps - this is built for exactly that.
Happy to answer any questions about how it works. I'm here all day.
Replies
Manual edits and proper integrations with the other players on the market are a must IMV - I don't want to spend time trying to explain something to a LLM when I can do it faster and more precisely
@sk_uxpin the desighn system point stands out a lot. If the AI can’t fully respect existing components, it creates more work instead of saving time. That gap feels bigger than most people expected.
@alex_j_jemmy can't agree more!
Asa.team
The community inventing a cost mitigation strategy on day one (Opus for drafts, Sonnet for edits, Haiku for tweaks) is a pretty clear signal the pricing model and use case are misaligned.
How does Forge handle this, usage based or a different model?
@ng_junsheng we are a little unique because we are an existing design tool our users are already on a license pricing model.
So we've added inclusive prompts into each licence level for free, with the option to purchase more if they exceed their limits.
Licenses include our entire design suite of tools + Forge, with users being able to easily switch between manual and AI workflows.. So unlike Claude users - when they run out, they can still edit manually without delay.