If you're pushing AI to your team, they

have reasons to push back. For the first

time in 200 years, we have a technology

that watches your team in order to learn

how to replace them. This is not

paranoia. I run three companies with a

team of 140 people, and this is what I'm

seeing.

6 months ago, the conversation was AI

changes everything. Now, what I'm seeing

is people pushing back against it, and

it's coming from inside my own team.

Last week, one of my full stack software

engineer sent me this on Slack as a

response to my latest video on hand

coders being angry. The more you use AI,

the more you depend on it. If it gets

bad enough, it can lead to someone

letting a model do most of the thinking

for them. It's making people stupid, and

that is what is making people angry,

including me. That is someone using AI

daily to build software, experiencing

the discomfort in real time, and his

frustration is pointing at something

that nobody's explaining properly. So,

let me explain it.

You've heard the numbers. AI makes

engineers 10 times faster. AI doubles

your team's output. Most of that is just

vibes. Some of the numbers are real. The

best measured number I've seen is the

median productivity gain across hundreds

of software engineers, and that is 7.8%.

7.8%.

I guess it's better than nothing, but

it's a long way from 10x. I'm not

telling you AI is a dead end. I use it

daily across my three companies. I'm

just telling you what the measured

number actually is, and the operators

signing the checks are reading it in the

same way. One CTO told the Pragmatic

Engineer survey, quote, "It is hard to

keep our CFO supportive of investing in

these tools because the productivity

benefits have proven difficult to

conclusively prove. CFOs cannot prove

the gain, and the developers are feeling

that same number from their side." Of

the engineers who saw their peak AI gain

in one quarter, 66%

saw it drop in the next quarter. The

early win doesn't compound. it fades.

But the productivity boost isn't the

only thing AI is doing to us. Think

about how you get better at anything.

Repetition, struggle, the friction is

the training. When you outsource the

struggle to a model, you outsource the

thinking and the skill development. A

2025 MIT Media Lab study found exactly

that. Heavy AI use leads to measurable

cognitive atrophy in the people that use

it most. The researchers called it

cognitive debt. We're not just less

productive than the marketing says. We

might be eroding the skills that made us

valuable in the first place. So the

productivity gain is small, the skill

cost is real. And people are being told

their jobs depend on adopting it. This

is all pointing at something deeper.

Here's what nobody's clear about on why

people aren't on board with AI. For the

last 200 years, every major technology

has eventually made everyone richer. The

factory, the electricity, the car, the

internet. The owners always saw the gain

first, but the workers eventually caught

up. Wages went up, living standards went

up, the gain spread. But AI is the first

widespread technology where the owner

sees a productivity gain immediately,

and the worker often doesn't see a gain

at all. In many cases, the worker is

helping train the thing that will

eventually replace them. There's a video

doing the rounds that you might have

seen. Factory workers in some plants are

being asked to wear cameras while they

work. The cameras record every motion,

every decision. The data trains an AI

that will eventually replace them. These

workers are being paid to teach the

machine that will take their job. And

it's not just factory floors. Last year,

Mark Zuckerberg announced that Meta

would be putting recording software on

its engineers' computers, so the model

could learn from them. Weeks later, he

laid off 8,000 people. 8,000 engineers

cut frees roughly 4 billion a year.

Meta's AI spend this year is 135

billion. The maths tell you where the

bet is placed and it's not on the

people. That asymmetry has always

existed. What's new is the speed. In

1900, the factory worker saw the boss

get rich first and waited a generation

for their own share to follow. The

worker wearing a camera in 2026 sees

that gain go into their boss before

lunch. So when people say something

feels off about AI, they're not

imagining it. And here's where it gets

sharper. In March, Jensen Huang stood at

the stage at Nvidia GTC and told his

engineers that they should be spending

50% of their salary on AI tokens. Half

of their salary in addition to their

wages. Where do you think that money is

coming from? Not from end user revenue.

Quick one before I go on. If you find

this content useful, hit like,

subscribe, and give the video a hype. It

really helps.

The main issue people have with AI is

not this technology is bad. It's more my

boss is pushing this on me. They are

profiting from me using it and I'm not.

Look at how the layoffs are being

framed. Cloudflare CEO sent a memo

recently announcing he was laying off

20% of his workforce despite record

revenue. He called the people being laid

off the measurers. He labeled them as a

function and then he deleted that

function. You stop being a person and

you start being a job description that

AI can do. So people are asking

themselves a different question. Am I

not more valuable than ChatGPT with all

the judgment and experience that I've

built up? The data confirms what they're

feeling. A survey of 2,400 executives

and employees this year found 60% of

companies plan on lay off people who do

not adapt and embrace AI. The same

survey found that 48% of executives are

calling their own AI adoption a massive

disappointment. So people are being

forced into a tool on a threat of losing

their job where the return isn't clear

even to the people mandating it. Now,

zoom out. That's one extreme. Let's look

at the other. Uber rolled out Claude

code to 5,000 engineers and burned

through their entire 2026 AI budget in 4

months. Per engineer costs hit $2,000 a

month. And the AI companies that all

that money is flowing into, most of them

aren't profitable. Anthropic just

reported its first-ever profitable

quarter. It's the first time any major

AI lab hits profit. OpenAI isn't

expected to be profitable until 2029.

So, where is all this money actually

coming from? Nvidia recently agreed to

put a hundred billion dollars into

OpenAI, and OpenAI uses that money to

buy Nvidia chips. That's not a customer

relationship. That's Nvidia subsidizing

its own sales. Even Jensen Huang admits

it's fragile. In a leaked all-hands

meeting last November, the day after

Nvidia beat their earnings, he said, "If

we delivered a bad quarter, if we were

off by just a hair, if it just looked a

little bit quicky, the whole world would

have fallen apart." That is Nvidia's own

CEO. The whole world. Michael Burry, the

guy from The Big Short, says Nvidia is

like Cisco. Most people remember the

dot-com bubble as pets.com vaporware.

Burry's argument is that it was actually

Cisco. Cisco in 2000 was a real

profitable company whose equipment ran

the early internet. The bubble was that

the infrastructure got built years ahead

of demand. The technology was real. The

build-out was real. The market was just

way ahead of revenue. The demand for AI

is real. I see it inside my own

companies. The question is, how much of

it is organic and how much of it is it

Nvidia paying its own customers to buy

its own chips? So, the bubble is real,

but it's a financial bubble. The

technology underneath it is not. And

because we're so early in this, what AI

looks like inside my companies is

different from the marketing version.

Let me show you.

At We UC with our 20 engineers, the

productivity boost reality is probably

closer to seven to that 7.8% not the

10x. But I don't think this 7.8% is a

ceiling. I think it's the cost of being

early. Let me give you two examples. One

of how AI has worked for me and one how

it's not worked out well. A few months

ago, my coffee machine broke. It's an

integrated machine, complicated one.

Normally, I'd email the manufacturer,

wait for their response, probably call

them because they haven't responded to

my email, wait on hold, try to book a

service, get passed to different

departments, and eventually get through

to someone and and try and book a

service for somebody to come and see the

machine. The whole thing will probably

take me, I don't know, half a day's

work. Instead, I took a photo of the

machine that included the model number

and the serial number. I gave it to an

AI agent, in this case Perplexity

computer, and I told it what the problem

was and to go fix it. I genuinely

expected nothing to happen. The agent

read the model number from the photo,

found the manufacturer's service

partner, used my Gmail connector to

email them, waited to for replies, and

when they did, it kept going back and

forth communicating with them until it

came back to me with a list of

appointment options. I approved one of

them and a price, and the manufacturer

called me an hour later to confirm the

appointment. The engineer then turned

up, machine was fixed. End to end, they

probably spent five minutes on it. The

agent did the rest. Second example, and

this one is a bit messier. We built our

own AI code review bot. It reads every

merge request and it provides comments

through Claude. When it works well, it's

genuinely excellent. But last week it

ran 10 reviews and cost us about 100

quid. And on one of them, it recommended

to restrict its own permissions. Then on

the next run, it couldn't post comments

cuz it revoked its own access away. One

of my engineers described it as a

brilliant five-year-old with a a

doctorate and amnesia. The capability is

there, but the economics aren't quite

there yet. So, that's what AI actually

looks like up close in my world. One

clean win that saved me half a day, one

expensive bet still being tuned. The

7.8% is only cuz we're still learning.

This video sponsored by Palantir AI.

They do real-time translation that keeps

your voice. 60 plus languages translated

in under a second. Not a robot voice,

your cloned voice. And that's not a

small thing when you've grown up between

two languages like me. I'm half Spanish,

half English. I actually worked as a

translator when I was a teenager. So

this hits close to home for me. My dev

team is mostly based in Ukraine. Their

English is great, but when something

gets technical, they'll switch to

Ukrainian to explain it properly. And

Palantir handles that instantly both

ways. You can use it for online

meetings, broadcasts, in-person events,

or straight through their API. And

normally a human interpreter at a live

event would cost you about a grand a

day. This is a fraction of that, does

the job just as well, and doesn't get

tired. Scan the QR code or use the link

in the description and use my discount

code axelmollist30

to get 30% off and start your free 3-day

trial on me. Actually on them, but you

get the idea.

The people pushing back on AI, I get it.

They're not Luddites. They're seeing the

asymmetry. They're being asked to wear

cameras or have their screens recorded

so they can train the thing that's going

to replace them. And they're all feeling

the cognitive cost, the skill atrophy

that my own engineers are warning each

other about on Slack. And the people who

are deeply convinced that AI is going to

change everything, I get that too.

Because every time I see an agent do

something in a minute that used to take

me half a day to do, I see what they're

seeing. The technology is real. The

capability shift is real. It's just

smaller and slower than the marketing

version for now. Both things are true at

the same time. The bubble is real. The

technology is real. The frustration is

legitimate. The opportunity is

legitimate. There's a lot of AI noise,

but there far fewer real outcomes. How

are you seeing this play out? Is your

team resisting? And if they are, is it

the cognitive cost or the economic

asymmetry? This is what I talk about on

this channel. If this is the kind of

thing you want more of, hit like, hit

subscribe, give this video a hype.

Really helps. And if you want more of

this kind of thinking, the stuff that

doesn't fit in a video, what's actually

working inside my companies, what's

breaking, I share it in my newsletter

every week. Head over to axelmolisto.com

and enter your email to subscribe. Your

team's frustration is real. The

technology is not the issue. The issue

is the billions chasing what AI might do

tomorrow, not what it does today. And if

you've not seen what 6 months of AI

coding does to my dev team or did to my

dev team, that one's right here. See you

in the next one.