This article is inspired by a post from Tom Bennett who was equally inspired by a insightful posting from Thomas Ruck, please go read them both.

But that is not the point, or maybe it is!

I thought I would create an article that gives my take on the Apple paper that allegedly unveils the con that is AI. I thought I was going to explain all the technical detail in easy to understand distilled insights. But I would be addressing the pawn, and not the chess game.

And the legitimacy of Apple’s paper “Illusion of Thinking” is scuppered by the fact it was a pawn in a marketing ploy.

I read the paper, it is a solid empirical work that confirms what many suspected:

current LRMs are sophisticated pattern matchers, not true reasoners.

However, the paper oversells its findings. The puzzles test a narrow slice of intelligence, and the failure modes might say more about training objectives than fundamental limitations.

BUT I must say, the decreasing thinking tokens phenomenon is the paper’s most interesting contribution and deserves follow-up work, watch this space.

So to be clear, we’re seeing the limitations of current thinking approaches, not proof that machine reasoning is illusory.

Let’s Look At The Chess Board Instead of the Pawn

I cannot lie, I was worried about Apple (still concerned) but many of us thought Apple lost the AI race.

But what if the real game is rent-seeking? OpenAI burns billions building models. Apple charges 30% on every AI-powered app transaction. The incentive isn’t to win AI, it’s to own the tollbooth.

Who’s really behind when you’re collecting from everyone else’s innovation?

Maximize imageEdit imageDelete image

From Thomas Ruck’s post

The Cobra Effect in Cupertino: Why Apple’s AI Strategy Makes Perfect Economic Sense

While tech Twitter erupts over Apple’s apparent AI incompetence, a dismal WWDC, a paper seemingly designed to discredit the very technology they’re supposedly pursuing, a different story emerges when, as my friend Mark Scott likes to say, you follow the money rather than the mockery.

The first peculiar data point I came across:

Apple researchers publish a paper titled “The Illusion of Thinking” just as their company appears to have given up on competing in AI.

The conventional wisdom?

Sour grapes from a fallen giant. But behavioural economics suggests something more intriguing:

What if this is exactly what a rational actor would do?

Minimize imageEdit imageDelete image

Poorly timed and premature paper? Or marketing genius?

The Principal-Agent Problem, Cupertino Style

As disappoint as it is, the moaning Murtles are missing the point:

Apple’s shareholders don’t need Apple to build the best AI. They need Apple to maximise returns on AI.

These are radically different objectives with radically different incentive structures. OpenAI, Anthropic, and others are locked in an arms race that burns through capital faster than a Silicon Valley startup burns through runway. The latest models require hundreds of millions in compute costs. The return? Unclear. The path to profitability? Even less clear.

Maximize imageEdit imageDelete image

Courtesy of Thomas Ruck

Meanwhile, Apple sits on a different perch entirely. With the Foundation Models framework, they’ve created what my university economics professor would call “two-sided market”.

• Developers get free AI capabilities,
• users get AI-enhanced apps,
• and Apple? Apple gets 30% of every transaction.

Nobody Talks About The Numbers

Consider the revealed preferences here. If Apple truly believed AI models were the future, they’d be pouring billions into compute infrastructure. Instead, they’re investing in something far more lucrative:

transaction infrastructure.

Every AI startup is optimising for the same metric model performance.

But what if that’s the wrong scoreboard? Apple’s optimising for a different game: marginal revenue per AI interaction.

When ChatGPT processes a query, OpenAI loses money. When an iOS app uses Apple’s AI framework to enhance a premium feature, Apple makes money.

It’s The Same Old Apple Play: The Obvious Substitute Effect Nobody Sees

Here’s where it gets interesting. Apple isn’t substituting their AI for OpenAI’s. They’re substituting ownership for innovation. It’s the same playbook they’ve run before:

• They didn’t invent MP3 players; they owned digital music distribution
• They didn’t invent smartphones; they owned the app economy
• They won’t invent AGI; they’ll own the consumer AI transaction layer

The “Illusion of Thinking” paper suddenly makes sense. It’s not sour grapes, it’s price anchoring. A business play. I feel sorry for the researchers, because they’ve been played. A pawn in business. The title is catchy but premature from a technology architecture perspective. We’re seeing the limitations of current thinking approaches, not proof that machine reasoning is illusory.

Maximize imageEdit imageDelete image

Great image from Thomas Ruck

But that is the thing. It was only meant to be catchy. Marketing got the title to be so.

And so by questioning whether models truly “reason,” Apple reduces the perceived value of raw model capabilities while increasing the relative value of integration and user experience.

The paper never once gaslights our experience of the effectiveness of the “reasoning” prowess in these models. I prove it everyday in my. Call it reasoning, or call it probabilistic pattern matching, it is very effective.

Maximize imageEdit imageDelete image

The nuance has gone way over all our heads. Thanks to Apple’s marketing team.

The Perverse Incentive Structure

The hidden incentive here is beautifully perverse: Apple profits most when AI becomes commoditised. Every breakthrough by OpenAI or Anthropic that gets integrated into Apple’s framework increases Apple’s platform value without increasing their costs.

It’s like being a casino.

You don’t need to be good at gambling when you’re the house.

Maximize imageEdit imageDelete image

The Unintended Twist Of This All

By making AI integration free and frictionless for developers, Apple might accidentally accelerate AI adoption faster than any technical breakthrough could.

The barrier to AI implementation isn’t model quality: it’s integration complexity.

A mediocre AI that reaches a billion users through existing apps creates more economic value than a brilliant AI that requires a computer science degree to implement.

The Black Market Parallel

A way to ponder this is: In countries with complex regulations, black markets emerge not because people want to break the law, but because the legal path is too complex. Similarly, developers aren’t using AI not because it’s not good enough (because it is), but because it’s too hard to implement.

Apple just became the AI dealer who makes the complicated simple, something Google has failed at, and takes a cut of every transaction.

The Point, In The Data, That You Should Care About

Here’s the number that should terrify Apple’s competitors: There are 34 million registered Apple developers. If even 1% integrate AI features that generate 10/month in additional revenue per user, Apple captures 1 billion annually in pure platform fees—without training a single model.

But that’s just the first order effect. The real value, is the “compounding moat” is the lock-in flywheel this creates. Every AI feature that delights a user increases their switching cost to Android.

Every subscription started for an AI-powered app extends customer lifetime value. Every workflow that depends on on-device AI becomes another reason to upgrade to the newest iPhone.

Apple isn’t just capturing 30% of AI app revenue, they’re using AI to increase hardware upgrade cycles from 3 years to 2.5 years, adding $50B in annual hardware revenue. They’re reducing churn to competing platforms by another 2-3%, worth $30B in retained lifetime value.

Maximize imageEdit imageDelete image

The platform fees are nice, but the real prize is making their $1,500+ per customer lifetime value even stickier and AI features are the perfect glue.

The Bottom Line

Everyone’s asking why Apple can’t make Siri work. But that’s like asking why the NYSE can’t pick stocks. They’re not in the prediction business—they’re in the transaction business.

The real question isn’t whether Apple is behind in AI. It’s whether being “behind” is the most profitable position in the race. When everyone else is burning capital to build better models, Apple’s building the tollbooth.

And if history is any guide, the house always wins.

Before You Run Off…The Twist

I love the contrarian view and powerful but well hidden nuance. So here is a twist.

The tollbooth theory assumes AI will flow through apps like music through iTunes. But what happens when AI becomes the OS itself?

Apple’s betting yesterday’s distribution moat works tomorrow. The real players aren’t building models or platforms, they’re building the infrastructure that makes both obsolete.

Let’s not get too carried away applauding Apple’s ‘tollbooth’ genius. Nobody’s asking why OpenAI would keep paying tolls. The moment AI margins turn positive, every major player builds their own highway.

What’s Apple’s moat when compute costs crater and open source catches up?

Technical Capability ≠ Market Reality (Basically Twisted Again)

I say this to my team all the time of whom are mostly engineers and technical consultants:

Forced legitimacy is way more powerful than better technology architecture.

Netflix was going to build phones, Spotify would bypass app stores, Epic would create an OS.

Technical capability ≠ market reality. Apple’s moat isn’t the 30% toll, it’s that they own the customer relationship at zero acquisition cost.

OpenAI spends $50-100 per user to maybe get a credit card. Apple already has a billion credit cards on file.

It hurts everytime I tell my architect team, but better technology rarely wins, better distribution usually wins, better unit economics always wins.

Apple has all three. That’s why OpenAI pays the toll instead of building highways, actions reveal preferences, and their actions say the tollbooth is worth every penny.

Let me plug my business Eclipse AI Consulting. If your enterprise or organisation is not only looking for an expert in AI implementation and strategy, but also one with a vision and an advisory edge that gives your business the intelligence X-factor, then look no further.

This is not our first rodeo and our experience and skill runs wide and deep. Reach out to me or my team, and let us show you what value we can bring you.