Claude built me a perfect mockup.
I gave it a detailed prompt — color palette, layout direction, component hierarchy — and it returned something I'd genuinely be proud to show a client. Clean. Professional. Closer to my vision than I expected on the first pass.
I sent it to my client.
Their reply: "Love it! Can you make it more modern? And maybe tweak the layout a bit?"
The Medium Problem
Here's the thing about that feedback: it was completely accurate.
My client knew exactly what she wanted. She opened the mockup, looked at it for thirty seconds, and her eye immediately caught what was off — the nav weight, the section padding, something about the hero. She had a specific visual reaction to a specific visual thing.
The problem wasn't her feedback. The problem was the channel.
She opened a Slack message and tried to translate a visual instinct into words. And words are a lossy medium for spatial, aesthetic information. "More modern." "Tweak the layout a bit." These aren't vague — they're compressed. She knew what she meant. The translation just dropped half the signal.
What Happens When You Remove the Translation Step
I started giving clients a different kind of access.
Instead of asking for written feedback, I'd open the rendered HTML in Kevra and send them a link. They could click on any element — a button, a section, a nav item — and drop a comment directly on it. No formatting. No re-explaining what they were looking at. Just: this → this is what I mean.
What happened next surprised me.
The clients who I thought were "hard to get feedback from" turned out to be extremely precise. They circled the nav and wrote "too heavy." They clicked into the hero section and wrote "needs air." They pointed at the CTA button and wrote "this should pop more." Three comments. Fifteen words. Completely actionable.
They weren't imprecise people. They were precise people using an imprecise tool.
When you give someone the right tool — one that matches how they actually think about visual work — they'll show you exactly what they want. Because they always knew. They just couldn't say it in a Google Doc.
The Real Bottleneck in AI-Assisted Design
Everyone talks about improving their prompting skills. How to write better, more specific prompts. How to give AI better context.
That's worth doing. But there's a ceiling on how well anyone — developer, designer, client — can describe visual intent in text.
The real bottleneck isn't skill. It's medium. Forcing visual feedback through a text channel is like asking someone to describe a painting over the phone and expecting the listener to reproduce it.
When you remove the translation step, something changes. The designer who "just couldn't explain it" now drives the whole revision cycle directly. The client who gave you three rounds of vague edits nails it in one annotated pass. The PM who always felt like a passenger suddenly has a direct line to the work.
AI didn't get smarter. The people around you finally had the right medium.
Claude can generate the perfect mockup. The hard part was never prompting — it was giving everyone on your team a way to say exactly what they see.
Kevra is the collaboration layer for human + AI teams — a shared space where your whole team can review, comment on, and improve AI-generated output. Join the waitlist →
Stay in the loop
Join the waitlist to get early access and product updates from Kevra.