As you already know, I've been saying

the model itself is turning into a

swappable part. You just rent it, you

route it, and you move on. So, if models

keep getting cheaper, the big labs

should be stuck in a price war right now

watching their margin bleed out, right?

However, they did the opposite.

Instead of selling a model and walking

away, OpenAI and Anthropic reportedly

built teams that sit inside a customer's

business and ship a working system. The

industry borrowed the name from Palantir

and calls them forward deployed

engineers. So, what that means in plain

English is that they send their best

people into your office to build that

thing, not just to demo it. So, think

about how strange that is. The companies

that make the models decided the model

was the cheap part of the job. MIT

looked at the pile of company AI pilots

and demo and found something like 95% of

them produced no result you could

measure. Isn't that crazy? The model

worked fine, just [music] they just died

at the last mile, where somebody has to

put the thing inside the real job and

make it run on a Monday.

That old software deal where a company

hands you a login and wishes you luck is

dying with them right now. So, I'm

building a payroll and scheduling

accounting product right now. So, I'm

living this question, not theorizing

about it. The scariest thing was never

the intelligence. It was the person

willing to sit in your business until

the thing runs properly, which creates a

problem here because that is the one

thing you cannot check before you pay

for it.

Most of my revenue right now comes from

putting AI systems into real businesses,

and the demand [music] is real, like

people calling me real. But, a hot

market pulls in a crowd and not always

the good kind. There are teams who will

cash an owner's check and just disappear

once the demo looks impressive enough.

And the ugly part is that an owner

cannot inspect the system that does not

exist yet. So, the whole sales runs on a

slide deck and a confident voice. That

is not a small gap. That's just the gap

the 95% falls through. Now, obviously I

want to be fair because a course sells

you the exact same un-verifiable

promise. You buy the promise of a

result, only one or maybe two percent of

buyers ever get the result they want,

and nobody finds out which group they

landed in until the money is just gone,

right? I think that is why selling

courses started to feel bad to me. You

are handing someone a lottery ticket

[music] with your name printed on it,

right? Setting aside the work fixes that

though, because you own the outcome. It

runs or it does not, and both of you can

see which by Friday. So, you should only

sell what a client can inspect, and only

buy what you can watch running, which

leaves one question worth answering.

Which part of the business do you build

yourself into?

The labs go deep for a reason. Once you

are inside a customer's payroll data and

their weekly schedule, pulling you out

costs more than keeping you. Nobody

swaps out a six-month integration on a

Tuesday because a cheaper option just

showed up out of thin air. You know, my

take is that the boring operational core

is the safest place to stand as of now.

And that is why I go after payroll and

scheduling and accounting piece. The

cool, fresh stuff is for the other

geniuses who want it. I don't want it.

Think about your own business for a

second. You do not rip out the thing

that runs payroll and you do not wake up

excited to migrate your scheduling your

data again. Now, the labs won by going

one profession at a time. They started

with cybersecurity, then moved into

medicine, law, and whole lots of other

things right now. So, the lazy takeaway

that I want to kind of give out is

niching down, picking an industry, and

just be done with it. I do not fully buy

into [music] that. The trap is which

thing you niche. Is it the content

creation for roofers? I don't think it

makes sense to me. The content and

marketing are a stack of many different

skills. And honestly, most of us are

barely, barely scratching the surface of

any one of them, right? So, a bounded,

repeatable job is a different animal.

Payroll is payroll, onboarding is

onboarding, accounting is accounting,

scheduling is scheduling. These are

whole different things. And when the job

has clear edge and it repeats every

month, you can own it end to end. When

the job is a fuzzy pile of hand-creative

skills, you just simply cannot, and

neither can I.

So, it would be really bad for you to

watch where buyers are spending because

they spend in two places, and both of

them sit far away from the model, down

at the cheap infrastructure and the

routing, which was last week's whole

point, up inside the work itself, which

is this week's that I'm talking about.

The dangerous spot is the generic

middle. I use ChatGPT for my clients.

Just easier for your clients to do

without you every single month. That's

not something to brag about. So, my

whole take is, you know, you either go

big or go home, but with calculated risk

because your runway is real and one dumb

bet could end the game. So, sitting

inside the workflow only pays off when

the workflow is big enough to be worth

the wiring. A two-person shop doing five

jobs a week may not have the juice. Deep

integration ties your income to your

client's survival. Anyway, the model is

the cheap part now, and it gets [music]

cheaper every month. The boring seed you

built yourself into is the part that

nobody can copy. Think about that.