Three problems where I used AI tools to do work that would not otherwise have been done, and what changed as a result. In each one I set the rules and approved every decision; the AI did the build.
Repairing a 900-post product blog in two months instead of ten
Premium lubricants manufacturer · Technical Editor · 2026 · Claude Code, WordPress REST API, Google Search Console
The situation
A product blog built up over more than a decade had near-duplicate posts competing with each other in search, links pointing at retired product pages, chains of redirects, missing bylines and image descriptions, and product claims that no longer matched current technical literature. Fixing it by hand would have taken roughly ten months.
How I used AI
- Directed Claude Code to audit every published post through the WordPress REST API and measure each problem on the live site, not from exports.
- Wrote the editorial, trademark and claims rules the fixes had to follow, and kept a human approval gate on every product claim and every irreversible change.
- Checked product claims against the current technical data sheets and removed any figure the literature no longer supports.
The result
- About two months of work instead of an estimated ten. Every figure below was measured on the live site before and after.
- 31 clusters of near-duplicate posts (68 posts sharing 30,125 search clicks) merged or retired behind redirects; 120+ duplicate, thin or outdated posts retired in all.
- Posts with no internal links cut from 116 to 6; 16 topic hubs built.
- 102 redirect chains flattened to zero; the busiest redirect is 69% faster.
- Every one of 759 posts now carries a product card and a named byline, up from 374 and 336; 660 links moved off retired product pages.
- Image descriptions on 99.96% of 5,386 library images, after a sample showed 87% blank.
Rebuilding a legendary technical archive without changing a word
Vintage BMW Motorcycle Owners (VBMWMO) · 2026 · Claude Code, Python, Internet Archive
The situation
One of the most-used technical resources for vintage BMW motorcycles was a 212-page website written over decades by a single expert. It had no search and no structure, and years earlier a bad save had destroyed every special character on the site: fractions, diameter symbols, umlauts. The rule was absolute: his technical prose could not be rewritten, shortened or summarized, because a paraphrased torque value can put the wrong number in front of someone holding a wrench.
How I used AI
- Directed Claude Code to build structure around the text instead of changing it: a topic taxonomy, model tags, cross-links, search and an index of every specification.
- Recovered the lost characters from Internet Archive captures dating back to 2008, with a safety rule that accepts only the kind of character that was actually lost.
The result
- A searchable knowledge base indexing 3,387 specifications across 212 pages, with the author’s words untouched.
- 78 characters recovered across 21 pages; anything not recoverable is still shown honestly as damaged.
- Endorsed by the club’s president and vice president; launch is pending the board’s decision on the access model.
Cutting magazine layout time in half
Classic BMW Motorräder, VBMWMO · Editor-in-Chief · 2026 · Claude Code, Python, InDesign scripting
The situation
Laying out each 40-page quarterly issue by hand took more than 40 hours: placing member-submitted articles and photos, keeping the fixed pages and ads in position, and holding every page to the printer’s specifications.
How I used AI
Directed Claude Code to build a pipeline in Python and InDesign scripting that turns submitted content into a page plan and flows it into the magazine’s template, leaving me the editorial and design decisions.
The result
Layout time cut at least in half on the first run, from 40+ hours an issue to about 20, with a further halving expected as the process is refined over the next few issues.