¶ Notebook
Long-form notes.
Field notes: what broke, what paid off.
Where the longer pieces land: case studies, vendor breakdowns, strategy frameworks. What I learned running AI systems for real businesses.
- Ed. 23October 26, 2026 · 5-min read · Beginner
Your AI Strategy Is Probably Three Spreadsheets in a Trenchcoat
Most owners who say they have an AI strategy have a pile of tools with no named bottleneck. A strategy fixes one problem. A pile buys subscriptions.
Read →
- Ed. 22October 19, 2026 · 8-min read · Beginner
I Traced 30 Small-Business AI Statistics to Their Source. 20 Survived.
Twenty matched the primary document. Five were misquoted and five I could not find. The 91% revenue claim comes from a different survey than the one credited.
Read →
- Ed. 21October 15, 2026 · 4-min read · Practitioner
I Measured Every East Cobb Med Spa. Then I Ran the Same Checks on My Own Site.
Robots.txt passed, and four local queries still missed my site. The audit also found my markup says Atlanta while my page says Marietta.
Read →
- Ed. 20October 1, 2026 · 7-min read · Beginner
I Got Burned by AI Automation Twice. Here's the Checklist I Run Before I Trust the Next One.
Before you buy another AI automation, run five checks. They would have saved me the two I threw away.
Read →
- Ed. 19September 28, 2026 · 7-min read · Builder
I Shipped a 13-Agent Pipeline. Eight Agents Never Fired.
Multi-agent earns its keep on independent, parallel work. On one coherent change it just fragments the context.
Read →
- Ed. 18September 22, 2026 · 6-min read · Practitioner
What I Measure Now That I Ignored for My First Six Months
For six months I measured activity, speed, and install counts. Here are the four metrics that actually tell me an AI is working.
Read →
- Ed. 17September 19, 2026 · 8-min read · Practitioner
Ten SEO Beliefs I Held. I Measured Every One. Every One Was Wrong.
Ten things I believed about getting found, each with the number I measured against it, and the post where the finding lives.
Read →
- Ed. 16September 16, 2026 · 6-min read · Builder
I Installed 260 AI Skills. 150 Of Them Have Never Run Once.
I counted every skill, agent and MCP server on my machine against five months of session logs. Most of the roster has never fired.
Read →
- Ed. 15September 7, 2026 · 4-min read · Beginner
I Stopped Chasing the Newest AI Model. My Work Got More Reliable.
The newest AI model is not the best one for a task you already made reliable. Switching costs more than it saves.
Read →
- Ed. 14August 31, 2026 · 7-min read · Builder
I Kept Making the Context Window Bigger. The Answer Was Never in There.
Wrong RAG answers are usually a retrieval failure, not a context-window one. A bigger window hides bad retrieval behind more confident wrong answers.
Read →
- Ed. 12August 19, 2026 · 5-min read · Practitioner
I Audited My Own Backlinks. I Found One, And It Was Wrong.
Across two sites I could verify exactly one inbound link. It comes from a directory I never submitted to, it lists the wrong city, and it points at a URL that is not my canonical.
Read →
- Ed. 11August 19, 2026 · 6-min read · Practitioner
A Google Search Advocate Says Topical Authority Isn’t Real. The Leaked API Docs Disagree.
Google ships no topical authority score, and John Mueller has said the idea is just relevance renamed. The 2024 Content Warehouse leak lists siteFocusScore, siteRadius, and siteEmbedding. Both can be true.
Read →
- Ed. 10August 19, 2026 · 4-min read · Beginner
The First Automation That Paid For Itself (And Three That Didn’t)
The automation that pays off runs often, touches only what you control, and can be checked in a glance. The ones that break depend on someone else’s system.
Read →
- Ed. 09August 18, 2026 · 6-min read · Builder
Why I Stopped Trusting Eval Scores and Started Reading Transcripts
A green eval number is not proof your system works; I only found the real regressions by reading what the model actually said.
Read →
- Ed. 08August 7, 2026 · 5-min read · Practitioner
I Added AI to My Workflow and It Made Me Slower. Here’s How I Caught It.
Bolting an AI step onto a workflow can run fine and still cost you time, because now a human has to read, judge, and redo its output.
Read →
- Ed. 07August 6, 2026 · 6-min read · Practitioner
The Automation That Failed for Two Weeks Without Ever Throwing an Error
An automation can exit clean and still be broken. Here is how I caught one that quietly rotted for two weeks, and the check that made drift fail loud.
Read →
- Ed. 06July 31, 2026 · 5-min read · Practitioner
I Automated the Wrong Thing First. Twice.
Automation that runs clean can still starve your goal. Verify against the goal metric, not exit code 0.
Read →
- Ed. 05July 29, 2026 · 4-min read · Beginner
I Wasted a Year Trying to “Use AI” Before I Fixed One Boring Task
AI for your business isn’t a strategy. It’s one repeating task you hand to a narrow tool and check yourself.
Read →
- Ed. 04July 27, 2026 · 7-min read · Builder
I Built an Agent Framework, Ran It in Production, Then Deleted Most of It
I built an orchestration layer to run a multi-agent system, then found most of it never fired and it made every bug slower to trace, so I deleted it.
Read →
- Ed. 03July 23, 2026 · 7-min read · Builder
The Day My AI Agent Approved What I'd Already Rejected
Three independent reviews of one automation found seven bugs with zero overlap. One caught my own agent approving something I had already rejected.
Read →
- Ed. 02July 22, 2026 · 5-min read · Practitioner
The integration that looked done in the demo and broke in week two
AI demos pass the happy path. Week two tests the null field, the timeout, and the off-script question. Budget most of the work for after "it works."
Read →
- Ed. 01July 22, 2026 · 5-min read · Beginner
The AI tool I bought and never opened again
A tool you don't check isn't saving you time. It's hiding a problem you haven't found yet.
Read →