I keep a rough clock on my AI sessions. Not because I bill hourly — I stopped doing that years ago — but because the numbers are starting to get absurd enough that I want a record.
Yesterday was one of those days.
The Session
Five hours. That is how long I worked with Claude Code on Hawaii-Guide.com. The site had already been migrated from Expression Engine to static HTML earlier this year. The structure was clean. The content was not.
Twenty-three years of editorial content spread across 193 pages. Islands, attractions, beaches, restaurants, trails, FAQs — all of it accumulated over two decades of running the site. Some of it was current. A lot of it was not. And almost none of it had structured data.
So I sat down and started directing.
What Got Done
First, I had Claude evaluate the entire site and generate a list of SEO improvements. Not vague suggestions — specific, page-by-page recommendations for structured data that was missing.
Then we executed. Across those 193 pages:
- Added Place schema to every attraction, beach, and point of interest
- Added Restaurant schema with locations and operating details
- Embedded GPS coordinates for attractions across all islands
- Built FAQ schema and injected it into dozens of guide pages
- Performed contextual freshness audits on every island — Oahu, Maui, Kauai, Big Island, Lanai, Molokai — checking whether attractions had closed, trails had been rerouted, restaurants had shut down, and hours had changed
That last one is the part people underestimate. The schema markup is mechanical. Tedious, but mechanical. The freshness audit is research. You are cross-referencing current conditions against editorial content that may have been written five or ten years ago. For six islands. Hundreds of individual points of interest.
The Math
If Victoria had done this manually — and she has done similar audits before — the research phase alone would have taken two to three weeks. She would be googling every restaurant, checking every trail status, verifying hours and closures island by island. That is before a single edit gets made.
Then the editing. In Expression Engine, bulk content updates were a slog. No batch operations. No find-and-replace across templates. Every page opened individually, edited, saved, previewed. For 193 pages of schema markup plus content corrections, you are looking at another two to four weeks of CMS work.
Conservative total: four weeks. Realistic total: six to eight weeks of sustained effort.
I did it in five hours.
The Multiplier
The math is not complicated. A conservative four-week estimate is 160 working hours. Eight weeks is 320 hours. Divide either by five.
32x to 64x.
That is not a rounding error. That is not a marginal improvement. That is a full month of professional work compressed into a single afternoon session.
And this was not a synthetic benchmark. This was a real site with real content serving real visitors. The schema data is live. The freshness corrections are published. The GPS coordinates are embedded. Google is already recrawling the pages.
What This Actually Proves
People keep asking for proof that AI productivity gains are real. They want controlled studies and peer-reviewed papers. I understand the impulse. But the proof is sitting in my commit history.
193 pages. 107 schema implementations. Six islands audited for freshness. Five hours.
The bottleneck was never the AI. The bottleneck was always the human willingness to sit down and direct it with enough specificity to produce real output. The people getting 10x or 20x gains are not using AI harder. They are using it with clearer intent.
I have been building Hawaii-Guide.com for over two decades. I know the content. I know the structure. I know what is out of date and what is not. Claude Code handled the execution. I handled the judgment. That split is what produced the multiplier — not the model, not the tooling, but the combination of domain expertise and agentic execution.
The Protocol: AI does not replace expertise. It compresses the time between knowing what needs to be done and having it done. The multiplier scales with how clearly you can articulate the work. Five hours and 193 pages is not a stunt. It is Tuesday.
