Do Not Expect Great Success from AI

Do Not Expect Great Success from AI.

Put a little radically, there are two choices.

Actually control an AI company like OpenAI or Anthropic,

or simply play with AI.

By “play with it,” I mean exactly that.

Try a new model when it comes out,

write some code,

write some prose,

run strange experiments,

and spend the night trying things.

It is good.

It is fun.

I do it too.

AI may be one of the most entertaining toys humanity has ever made.

But one distinction matters here.

Using AI well and owning the economic power it creates are entirely different things.

AI’s Intelligence Is Not Yours

However smart GPT is, it is not mine.

However smart Claude is, it is not mine.

The model is not mine,

the weights are not mine,

and the computing infrastructure is not mine.

I do not set the price.

I do not decide which model launches when.

Nor when existing models are retired.

I simply pay a fee

for the right to access that intelligence.

The difference is larger than it seems.

Use AI all day, and its capabilities eventually begin to feel like your own.

You produce in minutes code that once took hours,

finish in hours research that once took days,

and handle alone work that previously required several people.

Several studies have already found that generative AI improves productivity on specific tasks.

A study of customer-support workers found that AI users resolved roughly 14% more issues per hour on average. An experiment with BCG consultants also found improved speed and output quality on tasks within AI’s capabilities.

There is therefore no reason to deny AI’s productivity effects.

The issue lies elsewhere.

Who has exclusive ownership of that productivity?

Productivity Available to Everyone Soon Becomes the Baseline

Suppose AI lets me work 30% faster.

Good.

My competitor cannot use AI.

Then that 30% is a competitive advantage.

But my competitor starts using the same model.

So does my customer.

So does the new hire.

AI features even become built into the product itself.

The situation changes.

My productivity has not disappeared.

I am still faster than before.

But the relative advantage disappears.

When cars first appeared, car owners were astonishingly fast.

Once everyone had cars, humans still traveled quickly.

But being able to drive was no longer a competitive advantage in itself.

AI is the same.

As AI grows more powerful, its value can increase.

At the same time, the ability to use it can become less scarce.

There is no contradiction.

Falling Costs Present the Same Problem

AI prices are falling rapidly.

According to the Stanford AI Index, inference costs for GPT-3.5-level performance fell more than 280-fold between late 2022 and late 2024.

An astonishing advance.

But as a business owner, you need to think one step further.

Am I the only one getting it 280 times cheaper?

No.

Competitors get it cheaply too.

New founders do too.

Large corporations do too.

My customers do too.

Falling AI prices therefore lower my costs

while potentially lowering the price of the ordinary intellectual labor I sell.

Ordinary writing.

Ordinary translation.

Ordinary design.

Ordinary code.

Ordinary market research.

When anyone can produce similar quality at very low cost, earning excess profits from that ability alone becomes difficult.

Higher productivity and competitive advantage are not the same thing.

The More Fundamental Issue Is Ownership

The more deeply you integrate AI services into your business, the more this matters.

The platform determines model pricing.

The platform determines which models are available.

The platform determines the API specification.

The platform determines when models are retired.

Both OpenAI and Anthropic continually retire existing models and replace them with new ones.

That is not wrongdoing.

It is something a platform can naturally do.

What matters is that users understand their own position accurately.

The platform decides; the user adapts.

If my company’s core functionality depends too heavily on a particular model’s characteristics, one model update can affect the entire product.

A price change alters the cost structure.

An API change requires system revisions.

If the platform makes my offering a built-in feature, my product’s differentiation may disappear.

Doing business on a platform and owning that platform are therefore entirely different things.

It Resembles a Gold Rush

The gold-rush analogy makes this easier to understand.

Some people became enormously wealthy by mining gold.

But as crowds rushed in, the more stable positions belonged to those who controlled bottlenecks such as land, railways, finance, equipment, and distribution.

AI has a similar structure.

Users consume tokens.

Platforms sell tokens.

Users adapt to new models.

Platforms launch new models.

Users redesign their work around models.

Platforms own the foundation on which countless users’ work runs.

More users can make competition fiercer,

but for the platform, the market itself grows.

The structurally advantageous position is clear.

Does That Make Every AI Startup Pointless?

No.

Taking the argument to that extreme would break its logic.

It is possible to build an enormous company using AI.

Companies such as Cursor and Harvey already show this.

But look closely at them, and an interesting point emerges.

Their value does not simply come from

“using AI better than others.”

That is not the source.

They have customers.

Brands.

Deep positions inside users’ workflows.

Distribution.

Data.

Product experience.

Enterprise contracts.

And as they grow, they seek greater control over their technology and infrastructure.

Even successful AI companies, as they grow,

convert AI proficiency into assets they can own.

This distinction matters.

You can succeed using AI.

But the reason for your success cannot remain “being good at AI” forever.

So There Is One Important Question

What remains if you remove AI?

Customers?

A brand?

A distribution network?

Data?

Contracts?

Cash flow?

An organization?

Technology?

An understanding of the market?

Would the business keep running if you switched to another model in a day?

If so, you are using AI properly.

If the disappearance of one AI removes your company’s competitive strength, however, reconsider whether you are using AI or depending on its platform.

The Core of the AI Era Is Ownership, More Than Intelligence

I believe the price of intelligence itself will likely keep falling.

Better models will emerge,

smaller models will become stronger,

open-weight models will improve,

inference costs will fall,

and AI will become standard in countless products.

Intelligence will become increasingly abundant.

What follows is economically interesting.

What remains scarce becomes more valuable relative to what has become abundant.

Customers.

Trust.

Distribution.

Capital.

Exclusive data.

Brands.

Physical infrastructure.

Regulatory standing.

Decision-making authority.

And relationships and positions others cannot easily replicate.

No amount of AI progress automatically creates these things.

As intelligence becomes cheaper, their relative importance may increase.

That Is Why You Should Not Expect Great Success from AI

Return to the beginning.

Do not expect great success from AI.

I mean: do not expect great success from proficiency in using AI alone.

If you want overwhelming economic power within the AI industry itself, the story is entirely different.

Control a foundation-model company such as OpenAI or Anthropic,

build a new foundation model,

control computing,

control distribution,

or make countless people use the intelligence you supply.

At that moment, your position reverses.

Instead of buying tokens,

others buy yours.

Instead of adapting to platform rules,

you set the rules.

If you intend to go that far, by all means pursue great success through AI itself.

Otherwise?

Just play with it.

Try new models,

write code,

write essays,

make strange things,

talk all night,

and enjoy using it whenever it speeds up work.

AI is a truly entertaining toy.

Just do not confuse playing well with a toy

with controlling the toy industry.

AI proficiency is not ownership.

Productivity is not competitive advantage.

Access is not control.

And ultimately, business’s most important question remains the same.

Not what can I use,

but what do I own?

Do not expect great success from AI.

Play with it to your heart’s content.

And if you truly want great success,

own something of your own outside AI.