How to turn your expertise into an AI audit agent — a case study

An AI agent trained on one expert's own framework can turn an hour of analysis into a first-pass draft in minutes, with the expert still reviewing and finishing every output in their own voice. That's the result from building an audit agent for a one-person design studio: audits that used to take about an hour now take minutes, with one recent one clocking 13 minutes from input to sent, with no edits needed.

Jaclyn Scales, the studio's founder, gives away a free homepage audit to win clients, but the offer only moved as fast as she could personally analyze a site, take notes, and record a walkthrough. The fix wasn't to replace her judgment — it was to take the audit out of her head and into an agent trained on her framework, her experience, and her standards. The agent writes a first-pass audit in her voice; she reviews it, tightens it, and records the final walkthrough herself. The bottleneck moved from "can I find the time to analyze this" to "can I review a draft that's already close."

The case study documenting this was itself built as a single, editable HTML file rather than a designed image or PDF — which meant Jaclyn could open it, fix any wording, drop in her own headshot, and send it back without a single round trip through a designer.

The pattern generalizes past design audits: any repetitive, expertise-heavy task that's capping an expert's calendar — the thing only they can do, that they wish they could clone — is a candidate for the same approach. Train an agent on the specific framework and standards already in someone's head, and let the human stay in the loop for review and final judgment rather than every step of production.