There's something genuinely fun about vibe-coding tools, even when you're not building anything serious with them. Watching an idea turn into something clickable in minutes is one of the better parts of my week as someone who's still finding their feet in design. It scratches that "I wonder if this would work" itch without needing to understand how the wiring underneath actually works yet. These tools are why I keep poking at ideas I'd otherwise jot down and forget.

Claude Design has been part of my prototyping tests and practice since it dropped a few weeks ago. Google Opal, on the other hand, was completely new territory for me - it's Google Labs' no-code AI app builder, and it had been on my recommendations list for a while now, so I finally wanted to see if it holds up to my current favorite vibe-coding, vibe-designing, vibe-prototyping tool (Claude Design, that is).

Building something I'd actually use

Finally fixing a mess I've been meaning to deal with

I wanted to build something I'd actually use. My bookmark folders are genuinely one of the most chaotic places in my digital life. So the app I decided to give a shot is a bookmark triage tool. You upload your browser's exported bookmarks file, the tool fetches each page, suggests categories, and lets me sort each one into Keep, Read Later, Archive, or Nuke. If it actually works, it'd solve a real problem for me.

The plan was to use the exact same prompt for both tools so the comparison would be fair. And to keep it specific where it mattered (like the four-button triage, the upload flow, and the color palette), but vague enough on layout and visual details that each tool could bring its own personality.

One technical thing worth flagging upfront though: for auto-categorization to work, the tool would need to fetch each bookmark and read what's on the page. CORS (browsers blocking cross-origin requests for security reasons) tends to get in the way, and vibe-coding tools usually run in browser sandboxes that don't love making outbound requests. So I went in knowing this might be a stumbling block.

Claude Design was my benchmark going in

And it didn't disappoint

Claude Design is Anthropic's prototyping tool, launched in April 2026 and still in research preview. It's a hybrid of vibe-coding and design - you describe what you want, it builds it, and you get a small set of editing controls to tweak the output afterward without reprompting.

My main worry going in was the web access situation, and it turned out fine - better than fine, actually. Claude Design handled the link parsing, but not in the way I assumed it would. Real cross-origin page fetches are blocked by the browser, so instead of actually visiting each page, it inferred the category and a one-line description from the URL and title alone, using the Claude API to do the reasoning. I could actually see this in the build notes; it was transparent about what it did and what it couldn't do. For URLs with descriptive paths, this works surprisingly well, but for something like a Medium article with a random slug, it'd be less accurate, but for my samples, the inferences were spot on.

The output was a complete working prototype with an upload screen, triage view, summary screen with category groupings, and an export option to download a fresh bookmarks HTML file. Aesthetically I liked it overall, it had a calm palette, generous whitespace, and nice tactile buttons. But the font choice wasn't my favorite. Since Claude Design has those editing controls, I could swap the font though, and also tweak a few small other things without reprompting, though I still can't drag elements around manually.

Google Opal was a bumpy ride

It took several attempts and a workaround

Google Opal works on a visual workflow system, so instead of just describing your app and getting code back, you get a node-based canvas where each step (input, AI processing, output) is its own editable block. But there's also a meta-prompt bar at the bottom where you can describe what you want, and Opal auto-generates the workflow nodes from that. The right panel switches between a Preview, a Console, and Step and Theme tabs for fine-tuning. It's still completely free during the beta.

Things started well. Opal built a workflow with four nodes - file upload, parse, analyze and categorize, and generate triage dashboard. Then everything stalled. I sat watching the "Thinking..." spinner for about 20 minutes before reloading, but even after the reload, I just got my bookmarks listed back at me as plain links, without any of the analysis or triage interface.

So I went shorter and simpler and started a fresh app with a shorter prompt. That run got further, but it interpreted "let me click through them one at a time" very literally - instead of building a self-contained interactive app, it ran the triage as a conversation inside Opal itself, surfacing one bookmark at a time in the side panel and asking me to pick Keep/Read Later/Archive/Nuke. That's because Opal is more of a workflow runner than an app builder. When you ask it to build an app, it builds a workflow that simulates one by talking to you step by step.

Once I switched to App view and worked through my bookmarks, though, the experience was actually solid. It had clean visuals, smooth interactions, and downloadable output. I'll also say I liked Opal's aesthetic more than Claude Design's. The deep blue interface felt more cohesive, the buttons were beautifully laid out, and the typography was cleaner. Not bad, but it took several attempts and a lot of patience to get there.

Reliable beats pretty

Claude Design did exactly what I expected, and I'm still impressed by it especially given it's still in research preview. Opal is the more interesting one to think about - the workflow approach has potential, the final output looked great, and it's free. But for actually getting from idea to working prototype without losing an afternoon, Claude Design is the one I'd open first.