This is a guide to putting AI to work inside a company without losing control of it, and it gets revised instead of reprinted. Twelve chapters, free, no account. Agents I built read the field every day. I decide what makes it in.

Why it exists

I set out to write a book called The Ultimate Guide to Agentic AI. Then I looked at how long a book takes to reach a shelf against how fast this field moves, and the whole idea collapsed. A book like that is finished roughly a year before anyone reads it. The best ones on the shelf right now were written before most of what they describe existed. The subject narrowed after that, from agents in general to the decisions a company actually has to make about them, which is the work I do and the only part I can write about honestly.

So this is the version that does not go out of print. The structure of the work changes slowly and the twelve chapters change slowly with it. The back matter changes constantly. Revising a chapter costs nothing. Reprinting a book costs a year.

How it is built

The guide runs on a five stage pipeline. Agents handle four of them.

Chapter four explains that architecture properly, and chapter seven explains how the knowledge behind it is stored and retrieved. If you want to know how the machinery works, those chapters are the answer and this site is the working example.

It is also a demonstration of itself

The guide argues for a way of building these systems. This is how the guide itself is built.

01 Monitor

Pull from the watched sources.

Ch 3 · this is an agent
02 Screen

Score against what the guide already says.

Ch 9 · evidence, not confidence
03 Draft

Write the revision for the affected chapter.

Ch 4 · these exact layers
04 Review

I read it. Nothing skips this.

Ch 6 · a human approves Ch 12 · rules in the machinery
05 Publish

Build, deploy, regenerate the audio.

The changelog

Agents run four of those. The one they cannot run is the one that decides what you see.

Its own register entry

Chapter twelve asks you to keep a register: one row for every automation that touches company data, with six columns. It would be poor form to ask that of a reader and not publish my own. This is the pipeline above, in the format the chapter specifies.

Register · Guide revision pipeline1 of 1
What it does

Reads published material about this field every day and drafts revisions to whichever chapter it affects.

Owner

Jeremy Quinn

What it can read

Public sources on a list I maintain, and the current text of the guide. Nothing else on the machine, and no account of mine.

What it can change

Draft files in one folder. It cannot publish, deploy, or send anything.

Data tier

Public. No internal or restricted material goes near it.

If it is wrong

A bad draft sits in the folder until I read it. The failure mode is my wasted time, not a wrong page. I would notice at the review step, which every revision passes through.

Last reviewed

10 August 2026

Seven lines, and they answer every question anyone has asked me about this thing. That is the argument for the register in chapter twelve, made at a scale of one.

I have not found another one like it

Living documents exist. Sites written by AI exist. What I have not seen is a structured guide on this subject whose maintenance is an agent pipeline, whose revision history is published next to the chapters, and whose own architecture is explained inside one of them.

If you know of another, I would like to see it. That is not a rhetorical flourish. This is a format I am figuring out in public and I would rather learn from someone who got further.

What is honest about it

Everything below is true and I would rather you hear it here than work it out yourself.

Who is writing this

Jeremy Quinn. Two decades in enterprise storage and infrastructure, including running the team that took FreeNAS to TrueNAS. These days I run agent systems in production: an always on operations agent on hardware I own, storage administration over MCP, a market analysis platform, and the pipeline that maintains this guide.

I work with companies on the same problems the chapters cover. Which work belongs to a machine and which does not. What an agent should be allowed to touch. How you know it worked. Whether anyone is still using it thirty days later.

Why the site is the resume. Anyone can say they build agent systems. Fewer people have one running in public where the output is dated, the sources are named, and the revision history shows what changed and when. The changelog on the home page is the pipeline's output. If it stops running, you will see that too. That is the point of publishing it this way.

Working together

I take a small number of engagements at a time, usually starting with a short assessment: what work is actually happening, where AI belongs, where it does not, and what each fix is worth. If that turns up something worth building, we build it.

If the guide is useful to you, or wrong somewhere, or you want to talk about the work, the address below reaches me.

jeremy@jeremyquinn.net