Key Facts
• I've integrated Claude and Claude Design into early ideation, replacing exploratory work I would previously have brought a junior designer in to do — faster, with more iterations, and earlier in the process than was practical before.
• Claude Design outputs HTML. That's changed how I hand off to engineering in ways that a Figma-first workflow never could.
• AI is not going to replace designers. The skills that will matter more are the ones that were always hard to teach.
This page doesn't have screens to show. The projects are under NDA and the work is ongoing. What I can talk about is how I actually use these tools, what's worked, what hasn't, and what I think it means for the discipline, because that's the more useful story anyway.
Will AI replace designers?
No. Not entirely.
The longer answer is that the skills that make a great designer aren't going away, they're just getting harder to hide behind "visual craft". The deep user knowledge, the ability to look at something and know immediately what's wrong with it and why, the judgment to know which problems are worth solving and which have already been solved... Those things matter more now, and the AI tools expose when a worker doesn't have these skills.
So what changes? The growth plan, especially for junior designers. A lot of the mechanical craft we used to spend years teaching junior designers — how to construct layouts, how to work within a system, how to iterate quickly — AI compresses that significantly. Which means we can spend that time teaching the things that were always more important but always harder to get to.
Where I use it
I use Claude in early ideation, specifically to generate options that stay within an existing design system and don't introduce novel interactions. That constraint is intentional. I'm not looking for invention at that stage, I'm looking for breadth. Several directions, quickly, so I can figure out which ones are worth the time.
That used to mean pulling in a junior designer. Now I can do it earlier, faster, and with more iterations before anything gets in front of a stakeholder.
The output is a starting point. That part doesn't change.
What I've learned so far
The thing I expected to work best was design system adherence, especially after Claude Design launched. It still doesn't stick the way I'd want it to. The tools will re-solve problems the system already has answers for, or use components in ways that are clearly off. That's a real limitation and it requires active oversight, not occasional check-ins.
What's surprised me is how useful early, rough output is as a research tool. When I'm not yet sure what a solution needs to include, I can get something workable in front of users fast enough that it actually changes what I learn. Users respond to something concrete in ways they don't respond to a question or a concept. Getting to that artifact quickly has changed how I approach early discovery.
With Claude Code, before Claude Design, the AI workflow pushed toward one person doing everything — PMing, designing, and building — with everyone else in a feedback role. That's fast, but it flattens the process in ways that aren't always good for the work or the team.
Claude Design outputs HTML. That means engineering gets something they can actually work with rather than a Figma file they have to interpret and rebuild from scratch. The design handoff is back, and it's more useful than it was before. Decisions get discussed and defended again. That's been a better process.
Owning decisions in crit
When you present work made with AI, there are decisions in it you didn't actively make. The AI chose something, you accepted it, and now you're in a review without a clear answer for why it is the way it is. And that changes how we should bring work to critique and what we should be investigating in critiques.
Getting good at this, auditing what the AI did, owning the decisions worth owning, changing the ones that aren't, is a skill that's becoming more important fast. It's a harder version of something designers have always needed: knowing the difference between a deliberate decision and something that just happened. AI makes that question come up more often and in front of more people. One easy way to start doing this is identifying early on and coming to crit with a list of which decisions you as the designer intentionally asked the AI to complete, versus not, as a way of helping focus yours and others attention.