Internal design comparison — not for publication as-is

Five AI harnesses redesigned the same report from the same brief. Here's what each one built.

Real scan data from a completed webcoda.com.au analysis (overall score 79, Grade B, 8 pages) was handed to five model/harness combinations with one brief: serve both the time-poor business owner and the advanced user without compromising either, and back every recommendation with scan data we actually have, no speculative AI analysis. Screenshots below are the unedited, first-pass output of each run.

Five AI harnesses converging on a report
Source scan ai-checker.webcoda.com.au/analysis/6447c792… Task redesign.html, self-contained, real data Runs 5

What all five agreed on, independently

Every single harness landed on the same core information-architecture decision without being told to: two explicit views (a business summary and a technical/advanced view) behind a toggle, rather than scattering expandable detail through one long page. Where they differed was in tone, visual hierarchy, and how transparently each showed its working, not in the underlying structural idea.

Current live report, for reference Current live AI Checker report, a single long scrolling page
GPT-5.6 Sol Claude Code
GPT-5.6 Sol on Claude Code redesign of the AI readiness report

Business summary / Technical detail toggle. Priority action framed as “Fix this first” with measured evidence (37, 36, 0/19).

View live redesign →
GPT-5.6 Sol Codex CLI (native)