## Short - Is it better than Anthropic's usual releases? — yes. - Is it a good product? — somewhat yes. - Will I use it? — rarely. - Would I buy a $200 sub for this? — hell no. - Would I pay $100 for this? — no. - Would I pay $20 for this? — also no, you can prompt it 2 times and hit the limit.
## Now let's go in depth After trying the new Anthropic tool, I have some words to say. I am a Full Stack developer who started with Front End, and I've always worked on creating designs myself in code, so having something that can create designs from prompts with no code and is highly configurable sounds pretty good. It's nothing new — there are a lot of similar tools: Figma Make, Google Stitch, v0, Lovable, Bolt.new, Magic Patterns, and Uizard, to name a few. But we know that Anthropic has more power to ship better results. ## The setup I have a $200 Max sub and I wanted to improve designs for 2 of my projects: my open-source [Better-Fullstack](https://better-fullstack.dev) (2 pages) and my portfolio (3 pages). Starting with Better-Fullstack, I pasted a link to my website and it started creating multiple copies of designs. It asked me tons of questions about what I prefer, which was super cool — it even had a bar for how many designs I want. ![Claude Design questionnaire panel — vibe selector, variant slider, imagery options](./images/claude-design/image-1.jpg) ## Where it broke down Designs were pretty good, with rich features to change lots of things via Tweaks. I liked the Brutalist one the most, so we got rid of all the other designs and decided to work on this version. The comments feature was great — we tackled the small things I hated and got rid of them. Then I wanted to add more, so I asked it to create 5 unique designs based on the main one it had made. It hallucinated, created 5 ugly designs (GPT-4 level), looked absolutely horrible, and **the model deleted the original design**. I asked it to bring the original back, and… I got a new design, not the old one. During all this, the model kept stopping and writing that it was interrupted, so I needed to prompt it 3 times to continue. ![Better-Fullstack brutalist landing variant generated by Claude Design](./images/claude-design/image-2.jpg) ## Portfolio: even worse It was worse for the Portfolio. I connected GitHub, shared the repo and my own URL and got very bad designs, nothing close to my originals, even though I asked to _improve_ the homepage, not redo it. It didn't pick up any of my assets either. ![Portfolio "Currently / Stack I lean on" layout with the Tweaks panel on the right](./images/claude-design/image-3.jpg) ## So what is it actually good for? I spent some of my time on this, got not great results, and was tired of playing around with it — and I almost hit the limits. I think the tool is nearly useless to developers for most use cases. Those designs, questions, and patterns exactly match the designs I can get from Claude Code with Opus and design skills. I've done this trick a lot: ask for multiple designs, proceed with one, try to improve it — it's exactly the same. The only thing Claude Design gives me is easier control over changes, but it also hallucinates more and, for some reason, is super expensive. ## The price problem Jumping to the conclusion: the tool is pretty good, but useless because of the price. I can create designs with Claude Code and spend 2% of my 5-hour limit to do the same, instead of using 30% on designs. If it were usable for $20, it'd be a great tool for most devs. ## The bigger problem — too much, too fast However, I have another thing to say: too much. Too many releases of tools Anthropic could push to the future. The current trend of the world and the company itself is "ship more" — and this gets us to the issue of quality. Most of the releases, tools, and features are not well built, and sometimes unusable. **Claude Code is Anthropic's main tool after the models, and it's in one of the worst states of all the tools they're offering.** This brings me to the thought that I'd like more people focusing on fixing one tool instead of trying to release every possible tool without focusing on any. Updates to designs will become smaller and smaller, users will stop using it and forget about it — as many did with other tools — and fewer people will be focused on this. I don't believe they can maintain all of these closed-source tools in one company when their main tool can't reach good UX. I very much hope they are fixing and rewriting Claude Code, since it's been silent for some time which usually means a big update is coming. We'll see how it goes. Good luck building!