Michael Griswold

A one-day seminar in Houston, for owners and the person who runs their operations

Run your company on AI agents. One day, in a room with me.

You do not need to know how to code. You bring a laptop and one job you are tired of doing yourself. You leave with that job running on its own, inside rules you wrote, and a person on your team who can keep it running.

Apply for a seatTen minutes. Nothing is paid until we have talked.
A boardroom table set with flags and people seated around it
From the rooms.

The facts

What it is
A one-day seminar, in person. You and the person who runs your operations sit in a room of twenty with me for a day.
Where
Houston, Texas.
Who it is for
Owners, founders and family-office principals who run a real operating company, and the one person who keeps it running.
Seats
Ten companies. Two seats each: you and your operator. Both of you come.
When
The date goes to accepted applicants first. Apply, and if it is a fit you will have it within a week.
What it costs
I tell you on the call, before you pay anything. The price covers both seats.
What to bring
A laptop you use for work. Before the day, we set it up with you on a short call, so nobody spends the morning installing things.

This is for you if

  • You run the place. Owner, founder, CEO, principal, managing partner.
  • Something in your business is already a file. A shared drive, spreadsheets you use, a customer list, accounting software, email. One is enough.
  • You can name the person who keeps things running when you are out. They come with you.
  • You can decide to spend on this without a committee.
  • You do not want to learn to code. You will not have to.

This is not for you if

  • You are a developer who wants to build agents for a living. There are better rooms for that.
  • You want it done for you. You will do it, with me next to you, so that you can do it again.
  • Everything in your business is on paper and there is nobody to name for the second seat.

What happens in the room

Every step is done together, on your own files, before we move to the next one. Nothing is a lecture.

  1. First hour

    Your laptop already works

    Coffee, names, and a look at what you brought. The setup was done before today, on a call with us. You open your laptop and it is ready.

  2. Morning

    Your agent does its first job in front of you

    You point it at one real folder of your business and give it one real task. It does the task while you watch. You see what it can do and what it cannot.

  3. Morning

    You write the one page that tells it how you work

    In your own words: what it may do on its own, and what it must ask you first. That page is the whole trick, and you leave with it.

  4. Midday

    You learn to check its work

    You check it the way you check a new hire. Three questions, every time. You never take "done" on its word.

  5. Afternoon

    You give it a memory and some rules

    A memory, so you never explain the same thing twice. Rules, so the same mistake can only cost you once.

  6. Afternoon

    One job goes on autopilot, and your operator takes the keys

    You pick one thing that today lives only in your head. Together we set it up to run without you starting it. Your operator learns to run it.

  7. Last hour

    What you take home

    The files you made, a checklist for the next thirty days, and a way to reach me.

What you walk out with

  • One job in your business that runs on its own, inside rules you wrote.
  • The one-page brief that tells your agent how you work and what it never does without asking you.
  • A memory it keeps, with real entries about your vendors, your clients and your decisions.
  • Standing rules, so the same mistake can only cost you once.
  • A person on your team who can keep it running: your operator, trained next to you.
  • A thirty-day checklist and a way to reach me when you get stuck.

Why believe me

This is the first time I am running this room, so there are no graduates to quote yet. What there is: my own company runs this way, every day, and the numbers were counted, not estimated.

42,400
commits of work my agents produced in the last twelve months, from one desk. Counted from my own machine on 2026-09-13.
1.1 million
words of rules and memory my agents read before they act. Counted the same day.
26
rule files, each written the day something went wrong. One of them is printed below.
14,096
downloads of a public tool built this way, read from GitHub on 2026-09-13.

Every number, with the date and the way it was counted: the numbers page.

One of the rules my agents follow

There are 26 of these. I wrote each one the day something went wrong, in the words I was using that day. This one is short. Read it and you will know what a rule looks like, because you will write your own in the room.

rules/minimal-diff-complete-task.mdwritten 2026-06-24

Minimal diff, complete task - two different axes, both binding

Owner directive, Michael, 2026-06-24, twice-furious - non-negotiable. The short form of this rule lives in AGENTS.md ยง How to work; what follows is the full version.

"Minimal" bounds the diff's SURFACE. It never bounds the TASK. These are two different axes and conflating them is the failure this rule exists to name.

  • Minimal surface means: no unrelated churn, and never a patch over a root cause. Touch what the task requires, not more.
  • Complete task means: instructed to do N things, every part reaches a real done state. "Optional but correct" means DO IT. Half-doing the work while burying the gap in your summary is forbidden - a summary that reads as "done" over work that is not done is itself a second failure on top of the first.

When the work hits a snag, FIX the snag - never revert or back out requested work to dodge it. A green build achieved by deleting what you were told to create is a failure, not a success dressed differently. Only the owner may authorize undoing requested work, or deferring work that genuinely shouldn't be done - and when they do, the reason goes on record, not just in the commit that quietly drops it.

Michael with a group at an event
Michael with two men at an event
Michael Griswold speaking at a microphone

I host a monthly room in Houston for accredited investors and family offices, and it sells out. A few frames from the rooms.

Portrait of Michael Griswold

Who is teaching

I am Michael Griswold. For fifteen years I built and ran companies before I touched software: demolition, construction, manufacturing, a dairy, private credit, family-office structuring, capital raising. Then I built a software company, and I run it on AI agents. The numbers above are what that looks like.

I am not a programmer by trade and I do not write the code. I direct the agents that do, the way you direct people. That is the job I teach, because it is the job you already know how to do.

The deal

What I ask of you: come for the whole day, with your operator, with your laptop set up on the call before, and do every step.

What you get: one job in your business running on its own by the end of the day. If you did the work and that is not true, I keep working with you and your operator at no charge until it is, or you get your money back. Your choice, in writing, before you pay.

I do not promise a revenue number or hours saved. There is no evidence for one yet, and I will not make one up.

How to apply

  1. 1

    Fill in the application.

    Ten minutes. Your name, your company, what you run, and who your operator is. Nothing is paid.

  2. 2

    I read it myself.

    Within two business days you hear back from me: a yes, a no with the reason, or a question.

  3. 3

    We talk for twenty minutes.

    You get the date and the price. You decide. If it is a yes, your seats are held and we book your setup call.

Apply for a seat

Questions people ask before they apply

Do I need to know how to code?

No. Nobody writes code. If you can write an email and open a folder, you can do everything in this room.

What is an AI agent, in plain words?

A program you give a job to. Not a chat box you type questions into. You point it at a folder of your real files, you tell it how you like things done, and it does the job and reports back. It can read, write, sort, draft, check and send. It works while you do something else.

Is this the same as ChatGPT?

No. ChatGPT answers a question and forgets you. An agent does a job, remembers how you like it done, and does it again next week without you asking.

What do I bring?

A laptop you use for work. Before the day we set it up with you on a short call. You do not install anything on your own.

Will I be lost?

No. Twenty people in the room, your operator next to you, and I am there the whole day. Every step is done together, on your own files, before we move to the next one.

Who else is in the room?

Nine other companies like yours: an owner and the person who runs their operations. No developers, no vendors.

Why does my operator have to come?

Because they are the one who will keep it running after you go back to your day. If only you learn it, it dies the first week you are busy.

What does it cost?

I tell you on the call, before you pay anything. One price for the company, both seats. Nothing is charged when you apply.

What if the date does not work for me?

Tell me on the call. There will be another room.

What happens after the day?

You go home with the files, a thirty-day checklist and my email. The one job you set up keeps running. What you do next is up to you.

What if it does not work for my business?

Then I keep working with you and your operator, at no charge, until one job is running on its own, or you get your money back. It is written below.

Something else? Email contact@michaelgriswold.com and I will answer it myself.