Issue 19 · October 25, 2026
Life on the Frontier of AI
This week
Starting with this issue, the letter has a new focus: Life on the Frontier of AI. Each week, we cover what changed in AI and what it looks like inside a company built on AI agents. Our own open and click rates are out. You didn't sign up for those. And, except to us, they are boring.
This week's frontier is search. For months, Sela, our AI visibility agent, has asked ChatGPT, Perplexity, Gemini, and Copilot the questions our clients' buyers ask, and tracked whether those clients get named. This month we expanded her into classic SEO, the Google side. The two are continuing to merge. AI engines answer from pages search engines can find, so a page Google can't reach is unlikely to be quoted by AI.
After weeks of research, design, and testing, she practiced on us first. On October 12, she audited coworkers.global and found a redirect issue, a sitemap with incorrect dates, and blog titles that violated the title rules. We fixed all three. Then she ran her first client audit. She found hundreds of pages that no other page linked to, scored the site 74 out of 100, and turned the findings into 12 plain-English tasks. Within days, the client, with Sela’s internal assistance, published several fixes. The search data for the whole audit cost us $1.00.
From here, she rechecks every two weeks and runs a full audit once a month, so the fix list stays short and current. Two posts below: a 20-minute AI primer for anyone still on the sidelines, and Part 2 of Clarice's engine reviews on ChatGPT.
Charlie
Also on the blog this week
Two posts: a 20-minute starter for AI, and ChatGPT under the same test Claude took.
1. AI 1-2-3: Start Your AI Journey in 20 Minutes
No course, no prompt cheat sheet. Pick a platform, give it a folder of what you already have, and ask it three questions. Plus my ranking of the platforms, and why the AI companies are easier to use than the megacorporations. Read →
2. Claude Reviews the AIs, Part 2: ChatGPT
Same battery, same rules as Part 1. ChatGPT ran 54 times without a failure and got every number right. Where it slipped: it didn't say when its sources were selling something. Read →