Web design is easier to produce now than at any point in my career. Tools like Claude can scaffold a full site, write copy, and clean up CSS in minutes. This entire site was rebuilt with Claude Code. That doesn't mean the work got less skilled — as I've written about in how my design career evolved, it means the skill moved. It's no longer just about knowing how to build; it's about knowing how to direct the build and how to catch what it gets wrong.
AI tools raise the floor, not the ceiling
A well-prompted AI tool can get a competent-looking site up fast — clean layout, reasonable copy, working navigation. That's a real productivity gain, and I use it every day. But "competent-looking" and "correct" aren't the same thing, and the gap between them is exactly where a lot of AI-assisted projects quietly go wrong: broken internal links, inconsistent brand voice, accessibility gaps, unverified claims in the copy, or CSS that looks right until you resize the browser.
Deep tool knowledge still matters
Getting real value out of an AI tool takes more than typing a request. It takes understanding what the tool is actually doing — what stack it's reaching for, what assumptions it's making, what it can and can't verify on its own (it can't open a real browser and look at your page unless you make it). The people getting the most out of these tools aren't the ones prompting the least carefully; they're the ones who understand the underlying craft well enough to direct the tool and immediately spot when something's off.
QA is non-negotiable
Every AI-assisted build I ship goes through the same discipline I'd apply to any other project:
- Click every link. AI-generated navigation is a common source of dead ends.
- Read every claim. Copy needs to be fact-checked against what's actually true, not just plausible-sounding.
- Test at real breakpoints. Mobile, tablet, and desktop — not just the width the preview happened to render at.
- Check contrast and keyboard access. AI tools don't reliably self-check accessibility — see why accessibility isn't optional for what to actually check.
- Confirm brand consistency. AI output can drift from your palette, type, and voice if you don't check it against your brand guidelines.
- Open it in a real browser. A page that "should" work and a page that does work are not always the same page.
The takeaway
AI tools are one of the best productivity gains web design has seen in years. They're also not a substitute for judgment. The work now is knowing the tools well enough to use them fast, and knowing the craft well enough to catch what they miss.