A conversation

Living With
Artificial Intelligence

A working operator's view of what AI actually does in 2026 — the opportunity, the fear, and what an educated person ought to do about both.

2026
a conversation in six parts
What we'll cover

Six short pieces, in order.

We'll start with the basics, work our way up to the new thing called agentic AI, and end with what to actually do about it.

i.
What AI is
A plain-English foundation — what's actually inside, how it learns, what it can and can't do.
ii.
From chatbot to agent
The new word everyone's using — agentic AI — and why it matters more than ChatGPT did.
iii.
How I use it
My businesses, the framework I've built, and how the agents actually work alongside me.
iv.
Five honest fears
Jobs, control, deepfakes, privacy, bias — the questions I get asked, answered honestly.
02
roadmap
Part one · where we are

Something changed
in late 2022.

For seventy years, artificial intelligence was a promise that kept missing its deadline. Then in November of 2022, a research lab released a chatbot called ChatGPT — and a hundred million people used it in the first two months.

That was the moment AI stopped being a thing in laboratories and started being a thing on your phone, in your inbox, on your doctor's screen, in your grandchildren's homework.

03
the moment
By the numbers

The scale, in three figures.

800m
People using ChatGPT each week, as of late 2025 — roughly the population of Europe.
$300b
Capital expenditure on AI infrastructure in 2025, more than the entire annual oil-and-gas exploration budget worldwide.
3×
Speedup in legal, medical, and engineering work measured in controlled studies of professionals using AI as an assistant.

Whether you welcome this or not, it is the largest technological deployment in human history, measured by adoption speed.

04
scale
Plain English

It's not a brain.
It's an extremely good pattern-matcher.

What it really does

An AI model is given an enormous body of text — billions of pages from books, articles, websites, conversations.

It learns the statistical patterns of how language works. Which words follow which. Which ideas tend to appear together.

When you ask it something, it produces the most plausible continuation. That's it. That is the whole trick.

What that looks like
You type: "The capital of France is"

It computes: "Paris" (97%)
......................"Lyon" (1%)
......................"located..." (1%)

It writes: "Paris."

Multiplied by trillions of examples — across language, code, reasoning, conversation — that's modern AI.

05
how it works
How it learns

From the world's writing to a useful assistant, in three steps.

Data
Books, articles, websites, code, conversations — a substantial fraction of everything humans have written publicly.
Training
Massive computers spend months learning the statistical patterns — what follows what, how ideas connect.
Refinement
Humans teach it manners — what's helpful, what's not, what's truthful, what to refuse.
Model
A single file, perhaps a trillion numbers, that runs on a server and answers your questions.

A trained AI model is essentially frozen knowledge — a snapshot of everything it learned, then sealed. Every conversation with it is the same model talking. It doesn't grow between conversations. It doesn't remember you.

06
training pipeline
What it does well

Four things it does remarkably well today.

Text
Writing, summarizing, translating, editing, answering. The closest analogy is a tireless, well-read research assistant.
Images
Generating photographs, illustrations, designs from a written description. Also describing what's in a picture you show it.
Code
Writing software in any programming language. The most economically significant capability — and the one closest to what I do.
Reasoning
Walking through a problem step by step. Not perfect — but startlingly good at the kind of thinking we previously thought only humans did.
07
four capabilities
Capability one · text

"Summarize this 40-page contract
and flag anything unusual."

P
You
Paste in 40-page lease agreement. "Summarize the key terms and flag anything I should be worried about."
AI
Assistant
Summary: 5-year lease, $7,400/mo base rent with 3% annual escalator, tenant pays utilities and common area maintenance, landlord responsible for structural repairs only.

Three items worth flagging:
1. Section 12.4 — personal guarantee survives lease termination. Unusual for commercial.
2. Section 8 — "permitted use" is narrow; expansion requires landlord approval not to be unreasonably withheld.
3. Section 19 — automatic 5-year renewal unless 180-day notice. Calendar this immediately.

Took 6 seconds. Would have taken a lawyer 90 minutes — and is something most of us would never have asked a lawyer to do at all.

08
example · text
Capability two · images

"Draw me
a watercolor
of a quiet
New England farmhouse
at dawn."

Twenty seconds later, you have an original image. Not a stock photo, not a Google search result — a new picture, generated from your sentence.

This is the capability with the most obvious creative upside and the most obvious misuse problem. We'll come back to that.

09
example · images
Capability three · code

"Write me a spreadsheet that tracks my rental properties."

P
You
"Build me a property-tracker spreadsheet. Address, tenant, rent, lease end-date. Highlight any lease ending in the next 90 days in red. Calculate total monthly income."
AI
Assistant
Done. Here is a working Excel file with seven columns, conditional formatting on the lease-end date, a total-income formula at the bottom, and a sample row to show the format. Download attached.

No programming knowledge required. The same technique builds simple websites, calculators, scripts to organize your email, anything where the answer is software.

10
example · code
A real limitation

But there's a ceiling.
A chatbot can only talk.

You
asks a question
AI
writes an answer
END
that's the whole thing
One question · one answer · no memory between conversations · no action in the world

A chatbot can tell you to call your insurance company. It cannot call your insurance company. It can explain how to book a flight. It cannot book the flight.

11
the ceiling
"

A chatbot can tell you what to do.
An agent can do it for you.

— and this is where things get interesting
12
interlude
02
Part two
What's new in 2026

What's an agent?

If you've heard the word "agentic AI" and weren't sure what it meant — this section is for you. It is the most important shift in AI since ChatGPT, and most people have no idea it has already happened.

The familiar shape

ChatGPT in the browser is a one-loop machine.

You type a question question AI Model predicts an answer answer Text on your screen THE PROCESS STOPS HERE

It's like sending a fax to a brilliant librarian. You ask, you wait, you receive an answer, and the librarian goes back to sleep — forgetting your question and your name. Each fax is a complete, isolated event.

13
one-loop
The new thing

An agent is a multi-loop machine that acts.

AI Model the planner 1. Plan what should I do next? 2. Act use a tool · do something 3. Observe what happened? did it work? 4. Repeat until the goal is achieved

The agent doesn't just answer your question. It plans a sequence of steps, takes actions using real tools, observes what happened, and tries again until the work is done.

14
multi-loop · the agent
The recipe

Every agent is made of four ingredients.

The Model
The brain. The same kind of model behind ChatGPT — but now it's the engine of something larger.
The Tools
The hands. Things the agent can actually do — send email, search the web, query a database, run code.
The Memory
The notebook. So it remembers across conversations — what worked, what didn't, what you prefer.
The Loop
The work ethic. The agent keeps going — planning, acting, checking — until the job is finished.

A chatbot has a model. An agent has all four.

15
four ingredients
Ingredient: tools

"Tools" is what makes an agent useful.

Each tool is a specific power we grant the agent — like giving an employee keys to specific rooms.

web_searchlook something up on the internet
send_emailactually send a message to a person
read_fileopen and read a document on disk
query_databaselook up a fact in a company system
run_codeexecute a program and use the result
book_appointmentactually schedule something on a calendar
charge_credit_cardprocess a payment (with limits)
add_a_new_tooland many, many more — this is the point
16
tools
Ingredient: memory

The difference between forgetting and remembering.

Chatbot

Starts every conversation a blank slate.

Doesn't remember your name. Doesn't remember last week's question. Doesn't remember the long context you spent twenty minutes explaining.

Each session begins at zero.

Agent

Has a notebook it can read and write.

Remembers your preferences. Remembers the project you've been working on. Remembers what worked last time and what didn't.

It can compound knowledge the way an employee does.

In the system I built for my own location, each agent has its own area-specific notebook. The iOS specialist remembers iOS lessons. The backend specialist remembers backend lessons. The knowledge compounds, every day, forever.

17
memory
Worked example

"Book me a flight to Chicago for next Tuesday's meeting."

step 01
Reads my calendar. Finds the meeting on Tuesday at 2pm in Chicago.
tool: calendar.read
step 02
Searches flights. Filters: arrives Tuesday before 1pm, my preferred airline, aisle seat.
tool: flights.search
step 03
Checks the weather forecast for that day. There's a snowstorm warning. It re-plans — looks for an alternative that arrives Monday night instead.
tool: weather.fetch
step 04
Checks hotel availability near the meeting location for the extra night.
tool: hotels.search
step 05
Asks me for approval. "Here's the plan, total cost $847. Shall I book it?"
tool: ask_human
step 06
Books the flight and hotel. Sends me the confirmation. Adds it to my calendar.
tools: flights.book · hotels.book

Notice steps 3 through 5. The agent changed its plan when new information came in. That is what a chatbot can't do.

18
worked example
Side by side

The honest difference between the two.

Chatbot

ChatGPT, Claude, Gemini in the browser
What it does
Answers questions in text.
Memory
None between conversations.
Tools
Just its trained knowledge.
Steps
One question → one answer.
Can take action?
No. It can only talk.
Best for
Quick questions. Drafting. Brainstorming.

Agent

The new shape of AI in 2026
What it does
Accomplishes goals in the real world.
Memory
Persistent. Learns over time.
Tools
Email, calendar, databases, APIs, code.
Steps
Plans, acts, checks, retries — until done.
Can take action?
Yes — with whatever permissions you grant.
Best for
Multi-step work. Operations. Anything ongoing.
19
side by side
How the industry is organized

The agentic stack — built in layers.

Every agent system is a layered stack. Each layer is a different industry.

Applications ↑ what users see
Agent Platforms our Managed Agent framework
Tools & Infrastructure databases · APIs · email · calendars
Foundation Models OpenAI · Anthropic · Google · Meta

Foundation models — the GPT-4s and Claudes of the world — are commodities. The interesting work, and the durable value, is at the layer where you turn the model into a working employee. That's where we operate.

20
the stack
03
Part three
What I'm building

Two companies.
One framework.

Now let me show you what all of this looks like when an ordinary executive — not a Silicon Valley engineer — actually builds with it.

A little about me

A working operator. Who writes code.

2025
Buildout
Construction begins on IMAGE Studios Sharon
NO AI
FEB 2026
Doors open
My location is live. Salon suites operating.
STILL NO AI
MAR 2026
I start coding
Software for my location. AI-first from day one.
THE PIVOT
2026
It spreads
IMAGE Corporate takes interest. Founded Cadence Advisers.
WHERE WE ARE

Nine months from "no AI" to a national franchise in conversation about deploying my software. This is how fast this technology moves.

22
about me
My business

IMAGE Studios
Sharon.

A single salon-suite building. I rent suites to independent beauty professionals — each running their own micro-business inside my walls.

Sharon, MA · Opened Feb 2026
23
my single location
Why I built this

Every question. One inbox. Mine.

Me 1 owner Pro Pro Pro Pro Pro Pro Pro Pro Pro Pro Pro Pro Many pros in one building every question flows to me
hundreds
of small questions every week
24
the bottleneck
What came next

I built it once. Now it travels.

MARCH 2026
My location
I write software to run IMAGE Studios Sharon.
EXTRACTED
THE PLATFORM
The agentic framework
Coordinator + specialists + memory + tools.
DEPLOYED TO
IMAGE Corporate
In talks · 150+ locations · franchise rollout
Cadence Advisers
My consulting practice · any operator · any industry

Build one good thing for yourself. Then let the work speak for you.

25
Cadence Advisers
The architecture

The Managed Agent
framework.

A small team of AI specialists, each focused on one part of the business, overseen by a coordinator who routes the work. Like a well-run office.

26
framework
The org chart

One coordinator. Five area specialists. One coherent team.

Peter
Operator · Reviewer
The Coordinator
Router · Reviewer · Memory keeper
iOS
The mobile app for owners
READ · WRITE
Backend
Data, billing, the engine room
READ · WRITE
Hub
What pros and franchisees see
READ · WRITE
Integrations
Email, CRM, third-party systems
READ · WRITE
Platform
Auth, deploy, anything cross-cutting
READ · WRITE

Each specialist has its own memory, its own tools, its own area of expertise. The Coordinator knows which specialist handles which kind of work.

27
the org chart
Meet the team · 1 of 2

The Coordinator — my chief of staff.

Reggie
The Coordinator
A single, persistent AI presence. Same name across sessions. Same memory. Same judgment. The point of contact for everything I want to do.
What Reggie does for me
  • i. Listens to me. I describe what I want done in plain English. No technical jargon required.
  • ii. Picks the right specialist. "That's an iOS task" or "that needs the backend." I don't have to think about it.
  • iii. Writes the assignment. Translates my intent into a precise brief that the specialist can execute against.
  • iv. Reviews the work. Checks every change before it goes live. Catches mistakes the specialist might miss.
  • v. Remembers everything. Every decision, every preference, every lesson — across thousands of conversations.
28
the coordinator
Meet the team · 2 of 2

A specialist — the team that does the work.

Hayes, Reese, Atlas...
The Specialists
Each specialist gets a unique name when it spins up. They work autonomously on their assignment, then hand the result back to the Coordinator for review.
A specialist's day
START
Picks up an assignment from Reggie's queue.
PLAN
Reads its memory for lessons from past similar work.
ACT
Writes the code, runs the query, sends the email, fixes the bug.
TEST
Verifies its work — runs the tests, checks the result.
HAND OFF
Returns the work to the Coordinator with a clear status. Reggie reviews and lets me know.
29
the specialists
The supervision rule

Two modes. Two levels of supervision.

The most important design decision in our framework. Every specialist operates in one of two modes — and the mode determines how much trust the system extends.

R

Read Mode

No state changes

What it does: investigates, retrieves patterns, answers questions, runs read-only diagnostics.

Supervision: none. Cost ceiling only.

When I use it: "How's the iOS app doing?" "What did we decide last month about X?" "Show me yesterday's errors."

W

Write Mode

Actually changes things

What it does: modifies code, deploys, sends emails, makes external API calls that change state.

Supervision: graduated. New types of work require my explicit approval. Routine work auto-approves once trust is earned.

When I use it: "Fix the bug." "Send the welcome email." "Deploy the new release."

30
two modes
In practice

How I work with my team — every day.

P
Peter
Operator
request
R
Reggie
Coordinator
assignment
S
Specialist
iOS / Backend / Hub / …
result
R
Reggie
Reviews
briefs me
P
Peter
Approves
Morning
"What happened overnight, and what should I look at first?"
Mid-day
"Ship the new pro-onboarding flow this afternoon, I'm presenting it tomorrow."
Evening
"Tell me about the three biggest decisions today. I'll review and we'll talk in the morning."
31
the workflow
An ordinary Tuesday

An ordinary Tuesday at my location.

07:42
A salon professional opens her phone and asks about a billing discrepancy on her last lease invoice.
Backend · read
07:43
She receives a complete, accurate answer with line items pulled live from my financial system.
— in 11 seconds
09:15
A maintenance ticket arrives — leaking faucet in suite 14, photo attached. Diagnosed automatically; vendor dispatched.
Integrations · write
11:02
An automation didn't fire when it should have. The platform specialist investigates, finds drift, files a fix for my review.
Platform · read & write
14:30
A new salon professional begins onboarding. The Hub specialist walks her through every form, every tour — patiently, every time.
Hub · read
22:11
After hours, a question arrives. Answered. Logged. Ready for me in the morning.
— always on
32
a day in the life
The shift

What this actually changes.

The old way

humans only · before agents
  • One owner, drowning in small operational questions
  • Knowledge lives in the owner's head — and only there
  • Every new professional onboards from a blank slate
  • Response time = whenever the owner is awake and free
  • Standards depend entirely on the owner's bandwidth that day
  • Scaling the business means hiring more humans, linearly

The new way

humans + agents · 2026
  • Owner focuses on what only a human can do
  • Knowledge accumulates in agent memory — never lost
  • Every new professional inherits everything the agents have learned
  • Response time measured in seconds, around the clock
  • Same standard applied to every interaction, by design
  • Scaling means giving agents more work — for free
33
old vs new
04
Part four
A practical interlude

What you
can do with this
today.

Even without building anything. Without writing a line of code. With just the free tools already on your phone.

For everyone in this room

Three things to try this week.

i.
Use it as a research partner.
Free tools — ChatGPT, Claude, Gemini — will answer almost any question you'd have once asked a librarian, a nurse, a tax preparer, or a contractor for a rough estimate. Treat it as a remarkably well-read assistant: useful, fallible, worth double-checking on anything important.
ii.
Use it to write.
Letters to the editor. Notes to grandchildren. The eulogy you've been putting off. The complaint to the insurance company. Dictate the gist; let it produce a first draft; edit until it sounds like you.
iii.
Use it to learn things.
Ask it to explain something you've always meant to understand — quantum mechanics, the bond market, how a transmission works, what your grandson actually does for a living. It will not condescend, and it will not get tired.
35
three uses
Try this tonight

Open ChatGPT. Paste this.
Watch what happens.

P
You
"I'm 72 years old. I have $400,000 in retirement savings, no debt, social security covers my basics. My granddaughter is starting college next year. What are three thoughtful things I could do with $30,000 of those savings — for her, for me, for both? Don't give me investment advice; give me ideas a thoughtful financial planner might raise."
AI
Assistant · in about 8 seconds
Three directions worth considering, each with very different tradeoffs.

1. A 529 contribution for her education. Tax-advantaged in most states, grows tax-free, can be transferred to other family members if she gets scholarships. Best if you value structure and the tax benefit...

[continues with options 2 and 3, plus the tradeoffs of each]

It will give you a thoughtful, useful answer. You'll have something specific to bring to a real conversation with a real advisor. That's the unlock.

36
try this tonight
05
Part five
The honest conversation

Five fears,
five honest answers.

No deflection. No corporate softening. These are the questions I get asked at dinner parties — and the answers I actually give.

Fear № 1 · Jobs

"Will AI take my grandchildren's jobs?"

Most exposed
Routine knowledge work
Paralegal research Basic accounting First-draft writing Customer service scripts Data entry Translation
Least exposed
Hands · presence · trust
Plumbers Nurses Teachers Executives Craftspeople Therapists

The person who replaces them won't be an AI. It will be another person, using AI.

38
fear № 1 · jobs
Fear № 2 · Control

"Will AI turn against us?"

NOT REAL
The Hollywood Fear
Robot uprising
No desires. No goals. No continuity. No interior life. They don't plot.
THE REAL ONE
The actual concern
Quiet abdication
We let AI decide what loan, what résumé, what news to show us — without understanding why.

The fear isn't a machine waking up. It's us going to sleep.

39
fear № 2 · control
Fear № 3 · Deepfakes

"Can I trust what I see and hear?"

A real person
REAL
vs.
CAN YOU TELL?
Made by AI
FAKE
WHAT CHANGED
Video stopped being self-authenticating evidence. What matters now is the chain of custody — who is showing it to you, and do you trust them.

If it makes you furious in under three seconds, that's the algorithm working — not the truth.

40
fear № 3 · deepfakes
Fear № 4 · Privacy

"What does this thing know about me?"

What you fear
The AI is watching me
Has a file on me
Remembers our chats
Listens through my phone
What actually is
Watching you for years
Your free apps Loyalty cards Browser cookies Your car Social media Your phone's OS

The privacy battle was lost in the 2010s. AI didn't change the rules — it just made them legible.

41
fear № 4 · privacy
Fear № 5 · Bias & Fairness

"Should AI decide about people?"

The problem
The black box
INPUT
Your résumé
OUTPUT
Denied
WHY?
— unknown —
The standard to insist on
The glass box
INPUT
Your résumé
OUTPUT
Denied
WHY?
Insufficient experience in role X — appealable

If you can't see why a decision was made, you can't fix it.

42
fear № 5 · bias
A closing thought

We have done
this before.

In 1900, a man in his sixties had lived through the railroad, the telegraph, the telephone, and electric light. By the time he died, he would see the automobile, the airplane, the radio, and motion pictures.

Every generation of educated people has faced a transformative technology and decided whether to engage with it or step aside. The ones who engaged — carefully, critically, but openly — got the better life.

I suspect the same is true now.

43
a closing thought
The floor is yours

Questions?

I'd rather have a good conversation than a polished monologue. Ask me anything — about IMAGE, about Cadence, about agents, about your grandchildren's careers, or about anything else.