Imagine a VPS configurator bleeding roughly a third of its users at the configuration step. The reason isn’t a mystery: too many decisions, and no shortcut for shoppers who already know what they want.

The old playbook for fixing something like that: write a brief, run a design sprint, bring in an agency, wait six weeks, spend a five-figure budget. Instead, I spent a week with Claude — and ended with five testable prototypes built on production data. (The numbers here are illustrative; the workflow is exactly what I ran.)

Day 1: build on real data, not assumptions

Through the Claude-in-Chrome extension, Claude read the live page DOM and captured every state across tabs — Plain OS, OS with Panel, Application, custom slider config. It pulled real prices across monthly, 1-year, and 2-year terms, and grabbed the brand’s canonical logo SVG straight from our CMS endpoint. The prototype started from production reality, not a designer’s guess at it.

Days 2–4: an honest critique loop

Then I iterated with raw feedback. “Sliders overlap.” “Fonts too small.” “Right rail should be empty on load.” “Trust badges in the wrong place.” Each round produced a working HTML file I could click through — not a static mock.

The biggest unlock was a shortcut for decided shoppers: “New to VPS? Start with a template.” Click “WordPress site” and the hardware and software auto-fill with green checks, then the location step expands for confirmation. One click, three decisions made.

The debugging loop that made it fast

When the prototype broke, I pasted the Chrome DevTools console output back to Claude. It read the stack trace, located the offending line, and shipped a four-character fix. The same browser that runs the prototype runs the debugger — no context-switch, no copy-pasting error logs into a separate chat. Claude saw exactly what I saw.

Not every conversion problem needs a six-week sprint. Some need an afternoon and an honest critique loop.

Where could you compress months into days?