Why AI Productivity Gains Don't Show Up in Company Results

CEOs are putting AI deep inside companies, and the company number still struggles to get past 20% productivity boost. Even where individual output genuinely triples. In this article we explore where the 3x actually goes: employees quietly pocket the hours it saved them, coordination costs don't compress, and whatever survives both gets competed away on price.

Sep 15, 2026

Why AI Productivity Gains Don't Show Up in Company Results

Sep 15, 2026
5 min

“Is the 3x productivity boost in the room with us right now?”

Cartoon of the promised 3x productivity gain never arriving at the P&L
The 3x, on its way to the P&L.

Three years of putting AI inside companies. Ours, and other people’s…

In three years I have not once seen productivity 3x at the company level. NOT ONCE. I have watched it triple at a desk for individual employees and sometimes across small teams. Then I’ve gone and looked at the company number and the company number struggles to move past 20%.

AI consultants (LIKE US) ALWAYS have explanations.

“The models aren’t ready.”

“The tooling is immature.”

“Change management.”

Every explanation points at the technology, and the technology is the only part of this that did its job. BULLSHIT!

Now, I’m going to ramble about where it goes, and somewhere in here you’ll recognise your own company.

I don’t have a fix for you at the end.

The second I hand you a fix I become the consultant I just insulted. So this is part one, and you need to feel the problem before anyone talks to you about solving it.

We started this because we wanted the money

Ionio in 2023, optimistic about AI implementation work
Us, 2023.

Before I start judging anybody, some context on the seat I’m writing from.

We got into this in 2023 because we believed it. Every company was going to need AI put inside it. We knew how to put it inside them. And yes, ofc, we wanted to be the ones getting paid for it.

And we got surprisingly far. We built genuinely good things. Some of it made our clients good money.

But the number everyone was quietly imagining, the one where the whole company moves the way the demo moved.

We went looking for the ark of the covenant…??? and we came back with a very good invoice.

For about a year now we’ve stopped taking “help us implement AI in our organisation” work altogether.

Everything that does work is somebody asking us to build a new thing instead.

So that’s the seat I’m writing from. I sell this. I believe in this. I cannot find the money…why??

Employees pocket the diff

The report that took a day takes an hour. Now watch the six hours.

Nobody finishes a hundred ad concepts at 11am and starts on the next nine hundred because “there’s still time in the day”.

Work expands to fill the time available. Parkinson, everybody knows it.

Nobody ever bothered writing down the reverse, because until now work had never contracted. The task shrank to two hours. The day stayed nine.

There’s a nastier layer under that. Showing you the surplus gets punished.

The guy who tells you the report takes an hour now gets three more reports at the same salary, or he watches the person next to him become surplus. Hiding it is self-defence.

I’ll go first. I finish my tasks 3x faster and then I chill. That is my honest answer and it’s the honest answer of almost everyone you employ.

The author, done with the day's work before lunch
Me at 11:04 AM

So here’s the part that never makes the case study. The 3x already got paid out. To your employees. In hours. Off the books, with no line in your accounts for it.

You bought the tool. They kept the dividend.

Fortune headline on the shadow AI economy and the MIT GenAI divide study

fortune.com/2025/08/19/shadow-ai-economy-mit-study-genai-divide-llm-chatbots

Read that screenshot slowly. The pilot failed and the 3x was already inside, running on personal accounts, and nobody thought to mention it to you.

Excerpt on employees using personal AI accounts at work

Nobody gets paid for the extra

Illustration of compensation structures that pay for hours, not output
Screenshot on incentives and unowned productivity surplus

You are not a saint either brother…

Look at what your comp structure actually buys. Hours. Presence. A title. Nowhere in your company is there a price on the output getting bigger.

So when the model doubles what someone can produce, you are asking them to donate the difference to you. People do not donate to their employer. I wouldn’t.

No KPI anywhere on earth says reinvest the hours AI just gave you back.

The surplus has no owner, and unowned money always disappears.

And it’s worse than that, because AI made your KPIs cheap to hit. People are Goodharting your metrics so hard.

The dashboard has never looked better for a company as slow as yours…

Under all the incentive stuff there’s a thing I’ve stopped being polite about. The model is now better than the median employee at the median task, and the median employee does not care enough to command it.

Call transcript screenshot
Pranav, on the definition of dumb

Pranav really said it better than anyone else could. He is very smart.

(bro added this part in himself)

Your org chart is the hole in the bucket

Chaotic depiction of an organisation three years into an AI rollout
Year three of the rollout. This is a painting of hell, by the way.

Say you fix both of those. Hungry people. Incentives that actually pay.

The company number still leaks.

Because a company is a machine for coordinating people, and coordination does not compress.

My free Tuesday afternoon cannot be handed to the designer who is blocked. Saved hours don’t transfer. They sit trapped inside individual calendars.

Speed up the part that was never the bottleneck and you have sped up nothing. Lol. LMAO even.

Then hierarchy takes its cut.

You didn’t fix the bucket. You gave everyone a bigger jug. Stupid.

The junior who automated the thing needs three approvals to ship it, and approver number three is the person the automation makes redundant.

When a mistake cost six months of engineering, three reviews were cheap insurance. Now the build costs an afternoon and the reviews cost three weeks. The gate is worth more than the thing it’s guarding.

Even if 3x productivity is real, it won’t make it to your P&L

Say some of it survives all three. Some does. It doesn’t stay.

Everyone got the same model, on the same day, for twenty dollars. Your competitor’s 3x landed the same Tuesday as yours and one of you was always going to bid lower.

I can price this one because it happened to me. Work I sold for $50K a month sells for $20K–$30K now. Same output. Better output. The market took the whole difference.

Everyone pays for Claude. Nobody pays for Notion AI.

I know exactly one man who pays for Notion AI and I think about him more than is healthy.

Enormous at one end, zero in the middle, and the middle is where every AI feature you own lives.

The only people I can find who definitely captured the 3x are the ones selling the models. It’s on their P&L. It has been the whole time.

Where did the rest of it go?? “I don’t know”

Confused executive looking at falling profits despite record productivity
You probably, looking at your profits go down even as the productivity is at all time high

I can account for maybe two thirds of it. Maybe.

Somewhere between the person and the organisation there’s leakage I still cannot name, and I would rather say that than build you a framework with five boxes in it.

Here’s how the conversation went, on god,

Pranav asked me on a call, “where it goes”. I said, “I don’t know”.

He asked again. I said I don’t know. Again.

Some of the tax is legitimate tho. Coordination is real work. Plenty of tasks were never the bottleneck, so tripling them was never going to move anything at the top. And 1.5x at the person level is, by any historical standard, an absurd gift. Most technologies never delivered a tenth of that.

And some of the bottom of the ladder is lying to you.

Study finding experienced developers were slower with AI while believing they were faster
Experienced devs. Slower, while certain they were faster.

People cannot measure their own 3x.

Some of what your team reported as tripling is vibes, and your case studies are built out of the vibes.

One more thing. For three years “the models weren’t good enough” was a real excuse and I used it too. It expired sometime in the last six months. Quietly, and at the worst possible moment, because the funding dried up in the same quarter and the prices went up — so half the companies that finally had a reason to do this had just finished deciding not to.

Part one of two

The “productivity boost” is inside your building right now. But it will never reach your pockets…

You can’t patch this with a tool. Every tool you buy gets taxed on the same inefficiencies by the same people.

So ask one question in your next review cycle, to every single person you handed an AI seat to this year. What did you do with the hours it saved you.

Then shut up and sit in the silence.

Part two whenever I have something honest to say about fixing it. Could be a while.

Create Fast — Impress Instantly

Transform Your Business with AI Solutions

Ready to scale with AI? Get a personalized strategy session with our AI experts. Discover how to automate, optimize, and accelerate your business growth.

BOOK A CALL

Book an AI consultation

Looking to build AI solutions? Let's chat.

Schedule your consultation today - this not a sales call, feel free to come prepared with your technical queries.

You'll be meeting Rohan Sawant, the Founder.
 Company
Book a Call

Let us help you.

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

Behind the Blog

Pranav Patel
Writer

Good boi. He is a good boi & does ML/AI. AI Lead.

Editor

Why AI Productivity Gains Don't Show Up in Company Results

Read Time:
7
minutes

“Is the 3x productivity boost in the room with us right now?”

Cartoon of the promised 3x productivity gain never arriving at the P&L
The 3x, on its way to the P&L.

Three years of putting AI inside companies. Ours, and other people’s…

In three years I have not once seen productivity 3x at the company level. NOT ONCE. I have watched it triple at a desk for individual employees and sometimes across small teams. Then I’ve gone and looked at the company number and the company number struggles to move past 20%.

AI consultants (LIKE US) ALWAYS have explanations.

“The models aren’t ready.”

“The tooling is immature.”

“Change management.”

Every explanation points at the technology, and the technology is the only part of this that did its job. BULLSHIT!

Now, I’m going to ramble about where it goes, and somewhere in here you’ll recognise your own company.

I don’t have a fix for you at the end.

The second I hand you a fix I become the consultant I just insulted. So this is part one, and you need to feel the problem before anyone talks to you about solving it.

We started this because we wanted the money

Ionio in 2023, optimistic about AI implementation work
Us, 2023.

Before I start judging anybody, some context on the seat I’m writing from.

We got into this in 2023 because we believed it. Every company was going to need AI put inside it. We knew how to put it inside them. And yes, ofc, we wanted to be the ones getting paid for it.

And we got surprisingly far. We built genuinely good things. Some of it made our clients good money.

But the number everyone was quietly imagining, the one where the whole company moves the way the demo moved.

We went looking for the ark of the covenant…??? and we came back with a very good invoice.

For about a year now we’ve stopped taking “help us implement AI in our organisation” work altogether.

Everything that does work is somebody asking us to build a new thing instead.

So that’s the seat I’m writing from. I sell this. I believe in this. I cannot find the money…why??

Employees pocket the diff

The report that took a day takes an hour. Now watch the six hours.

Nobody finishes a hundred ad concepts at 11am and starts on the next nine hundred because “there’s still time in the day”.

Work expands to fill the time available. Parkinson, everybody knows it.

Nobody ever bothered writing down the reverse, because until now work had never contracted. The task shrank to two hours. The day stayed nine.

There’s a nastier layer under that. Showing you the surplus gets punished.

The guy who tells you the report takes an hour now gets three more reports at the same salary, or he watches the person next to him become surplus. Hiding it is self-defence.

I’ll go first. I finish my tasks 3x faster and then I chill. That is my honest answer and it’s the honest answer of almost everyone you employ.

The author, done with the day's work before lunch
Me at 11:04 AM

So here’s the part that never makes the case study. The 3x already got paid out. To your employees. In hours. Off the books, with no line in your accounts for it.

You bought the tool. They kept the dividend.

Fortune headline on the shadow AI economy and the MIT GenAI divide study

fortune.com/2025/08/19/shadow-ai-economy-mit-study-genai-divide-llm-chatbots

Read that screenshot slowly. The pilot failed and the 3x was already inside, running on personal accounts, and nobody thought to mention it to you.

Excerpt on employees using personal AI accounts at work

Nobody gets paid for the extra

Illustration of compensation structures that pay for hours, not output
Screenshot on incentives and unowned productivity surplus

You are not a saint either brother…

Look at what your comp structure actually buys. Hours. Presence. A title. Nowhere in your company is there a price on the output getting bigger.

So when the model doubles what someone can produce, you are asking them to donate the difference to you. People do not donate to their employer. I wouldn’t.

No KPI anywhere on earth says reinvest the hours AI just gave you back.

The surplus has no owner, and unowned money always disappears.

And it’s worse than that, because AI made your KPIs cheap to hit. People are Goodharting your metrics so hard.

The dashboard has never looked better for a company as slow as yours…

Under all the incentive stuff there’s a thing I’ve stopped being polite about. The model is now better than the median employee at the median task, and the median employee does not care enough to command it.

Call transcript screenshot
Pranav, on the definition of dumb

Pranav really said it better than anyone else could. He is very smart.

(bro added this part in himself)

Your org chart is the hole in the bucket

Chaotic depiction of an organisation three years into an AI rollout
Year three of the rollout. This is a painting of hell, by the way.

Say you fix both of those. Hungry people. Incentives that actually pay.

The company number still leaks.

Because a company is a machine for coordinating people, and coordination does not compress.

My free Tuesday afternoon cannot be handed to the designer who is blocked. Saved hours don’t transfer. They sit trapped inside individual calendars.

Speed up the part that was never the bottleneck and you have sped up nothing. Lol. LMAO even.

Then hierarchy takes its cut.

You didn’t fix the bucket. You gave everyone a bigger jug. Stupid.

The junior who automated the thing needs three approvals to ship it, and approver number three is the person the automation makes redundant.

When a mistake cost six months of engineering, three reviews were cheap insurance. Now the build costs an afternoon and the reviews cost three weeks. The gate is worth more than the thing it’s guarding.

Even if 3x productivity is real, it won’t make it to your P&L

Say some of it survives all three. Some does. It doesn’t stay.

Everyone got the same model, on the same day, for twenty dollars. Your competitor’s 3x landed the same Tuesday as yours and one of you was always going to bid lower.

I can price this one because it happened to me. Work I sold for $50K a month sells for $20K–$30K now. Same output. Better output. The market took the whole difference.

Everyone pays for Claude. Nobody pays for Notion AI.

I know exactly one man who pays for Notion AI and I think about him more than is healthy.

Enormous at one end, zero in the middle, and the middle is where every AI feature you own lives.

The only people I can find who definitely captured the 3x are the ones selling the models. It’s on their P&L. It has been the whole time.

Where did the rest of it go?? “I don’t know”

Confused executive looking at falling profits despite record productivity
You probably, looking at your profits go down even as the productivity is at all time high

I can account for maybe two thirds of it. Maybe.

Somewhere between the person and the organisation there’s leakage I still cannot name, and I would rather say that than build you a framework with five boxes in it.

Here’s how the conversation went, on god,

Pranav asked me on a call, “where it goes”. I said, “I don’t know”.

He asked again. I said I don’t know. Again.

Some of the tax is legitimate tho. Coordination is real work. Plenty of tasks were never the bottleneck, so tripling them was never going to move anything at the top. And 1.5x at the person level is, by any historical standard, an absurd gift. Most technologies never delivered a tenth of that.

And some of the bottom of the ladder is lying to you.

Study finding experienced developers were slower with AI while believing they were faster
Experienced devs. Slower, while certain they were faster.

People cannot measure their own 3x.

Some of what your team reported as tripling is vibes, and your case studies are built out of the vibes.

One more thing. For three years “the models weren’t good enough” was a real excuse and I used it too. It expired sometime in the last six months. Quietly, and at the worst possible moment, because the funding dried up in the same quarter and the prices went up — so half the companies that finally had a reason to do this had just finished deciding not to.

Part one of two

The “productivity boost” is inside your building right now. But it will never reach your pockets…

You can’t patch this with a tool. Every tool you buy gets taxed on the same inefficiencies by the same people.

So ask one question in your next review cycle, to every single person you handed an AI seat to this year. What did you do with the hours it saved you.

Then shut up and sit in the silence.

Part two whenever I have something honest to say about fixing it. Could be a while.