Agent architecture
Why AI agents hallucinate, and what to change in the context
Unreliable agent output often comes from how the context is structured rather than from the model. Three architectural patterns that make answers easier to check.
Free · By email · One idea per issue
The AI Operator Newsletter
Practical notes on AI automation, tools and business processes. Each issue is built around one idea you can try in your own work, without the hype.
One actionable idea per issue
Most AI newsletters are a list of news. This one is built around a single idea per issue that you can put to work: a tool, an automation recipe or a technical pattern, with enough detail to try it yourself.
Subscribe
Leave your first name and email address and the next issue will reach your inbox.
Every issue follows this format
A predictable structure, so you can scan quickly and read closely where it matters.
The big idea
One concept, strategy or technique explained in two or three paragraphs. This is the takeaway you can use right away.
Tool of the issue
One tool recommendation with use cases, an honest assessment and the vendor’s published price at the time of writing.
Quick recipe
A small automation you can set up in about 15 minutes, with tool links and step-by-step instructions.
From the books
An excerpt or an expanded idea from one of the AI Operator Series books, useful on its own.
Example topics
The kind of subject an issue goes into.
Business building
A no-code stack for a small AI practice
Which tools, workflows and automations a one-person practice can run without writing code, and where the limits of no-code start to show.
Architecture
MCP or custom APIs: when to use each
The Model Context Protocol is powerful, but not always the right choice. A decision framework for when to use MCP, when to keep custom APIs and when to combine both.
Last updated: October 2026