The Money Is In The Distribution

ben@benarmstead.co.uk (Benjamin Armstead) — Sat, 29 Aug 2026 00:00:00 +0000

Prelude

Prepare yourself for a dive into technology and markets over the past 100 years.

“The longer you can look back, the farther you can look forward.”

Winston Churchill

I write this from a place of neutrality. I have put the past 10 years a good chunk of my lifetime into becoming extremely well versed in computers and software.

However, one must not hold nostalgia when making an economic decision. You must take the best of what you have learnt and apply it to the new age.

The Money Is in the Distribution

Three years ago, a friend and I had an extremely enlightening conversation with a French businessman in a bar in Yogyakarta.

The man shipped furniture from Indonesia to hotels all around the world.

He kept telling us:

“The money is in the distribution.”

As I get older, the more I see how applicable this advice is across industries.

Throughout his intoxication, he made three great points:

Don’t be the one dealing with the end customer. They are a pain in the backside.
Don’t be the manufacturer. They are commoditised into razor-thin margins.
The connector and distributor holds the power, the leverage, and the responsibility of which you are rewarded for.

When you look at Silicon Valley, you can see the same pattern.

Google distributes information.

Airbnb distributes accommodation.

Meta distributes attention.

AWS distributes software.

Amazon distributes products.

Uber distributes transport.

The list goes on…

How Do Distributors Form?

The aforementioned are all “Information Age” companies.

Just as with the Industrial Age, the Information Age and now the Intelligence Age each chapter has consolidation and commoditisation phases as it draws to a close.

What we are seeing with LLMs producing software at the speed of light, although some may not want to believe it, is exactly that.

It’s the end of the Information Age and the commoditisation of what was once a craft.

These industries do not disappear overnight, or even contract too much. They simply fade into the background of human production.

The factories from the Industrial Age did not disappear. The engineers of the Atomic Age and Space Age did not vanish. The focus of the world simply changed.

Most of the problems that needed to be solved with the capabilities we had at the time were solved, and the marginal returns on innovation became low.

The economic appetite for that industry to outgrow others fades, and it simply becomes a commoditised pillar of human economic function.

That’s not by any means to say it’s no longer a good industry to be in, it’s just no longer the world’s highest-margin focus.

We can see this if we look at the margins across industries.

Industry	Typical net margin
Grocery / supermarkets	1–3%
Airlines	2–5%
Automotive manufacturing	4–8%
Construction	4–8%
Restaurants	3–10%
General retail	3–8%
Industrial manufacturing	6–12%
Pharmaceuticals	10–20%
Banking	15–25%
Software / SaaS (mature)	15–30%+
Semiconductors	15–30%
Luxury goods	15–25%

The Information Age, our last age clearly leads by a mile, as all industries did during their “age”. But it’s highly foreseeable that LLMs will bring down the margins of SaaS towards those of other industries.

For engineers who want to be on the cutting edge of human advancement, we need to recalibrate our thinking and our role.

A good direction three years ago is not necessarily a good direction now.

Society has made much of the important software it needs.

Yes, there will be niches for the next five years that want their specific use cases served.

But the light is fleeting.

Fear not.

A phoenix rises from the ashes.

Opportunity is a moment where technology and appetite meet a problem waiting to be solved.

You were unlikely to bring the next Google to market.

You do have a chance to do so for the Intelligence Age.

We have spent much of humanity’s combined IQ over the past 20 years figuring out how best to get data from A to B.

Just as, in the decades before that, we perfected figuring out how to get humans and physical goods from A to B.

Now it’s time to get intelligence from A to B.

That is: applied in the correct way, on the correct data, at the correct moment.

So What Is This New Technology We Have?

An LLM is an electronic representation of the average of human information, and works, very loosely, somewhat like a human brain.

It’s missing a few vital functions, however. It’s likely we will discover these in the future.

Its main limitations at the minute are:

Originality. It’s likely our current technology cannot truly do this without becoming nonsensical.
Reasoning. This is knowing that if I do X here, Y will happen elsewhere. It is still poor at performing this at a standard that would be expected of a skilled human, despite having access to a mapping of information far beyond what any individual human could hold. This will likely be where some of the biggest future breakthroughs happen.
Context. Human memory. It can have very good context in certain situations for example, a codebase but is still let down by its reasoning in this regard.

I’m going to explain this by breaking it down into the three problems we face.

1. Reasoning and Context

Reasoning and context, at the minute, must be solved by people.

Divide and conquer.

Split the problem into small tasks you would hand to a junior employee.

Break it down.

Give clear instructions.

Do not ask the AI to “run the company”.

Ask it to perform a specific task, with specific information, within specific boundaries.

2. Originality

This must also stay human.

We cannot move creativity into a machine that everyone else on the planet has access to.

If your competitive advantage is simply asking the same model everyone else can access to generate something, you are building the most commoditised service imaginable.

The machine can amplify originality.

It cannot be where your originality comes from.

3. Safeguards

Just as a human employee gets things wrong, we need to put the correct safeguards in place when implementing AI so that it doesn’t do something wrong.

Luckily, this is not a completely new problem.

It is much like writing tests in software, which we have done for decades.

Define what should happen.

Define what should never happen.

Test the output.

Add human approval on top.

Then allow the AI to earn the right, over time, to act autonomously.

How Do We Ensure We Are the Distributor in This New Age?

Don’t be the man making the model. i.e. the Indonesian making the hotel chair.
Don’t be the man talking to the end customer in every industry i.e. the insurance business, the marketing agency, etc., that you are selling to.
Be the connector. Be the one that links the companies selling to the end customer with the model. You are the one delivering the implementation and the intelligence.

When we look at it this way, we can see where the money lies.

It’s in the platform that others will be running their intelligence on.

It seems much more plausible to me that the harness is where the value will lie, alongside the support given to companies to get it implemented.

If a company uses you for its intelligence, you are the distributor of intelligence.

You are the winner.

You don’t have to sell to their end client and manage that relationship. You provide the intelligence that gets the job done for their end client.

You are the distributor.

The model itself may become increasingly commoditised.

The intelligence it contains may become increasingly cheap.

But getting the correct intelligence, into the correct business, against the correct context, with the correct safeguards, at the correct moment?

That is distribution.

Therefore, just as important as the platform used to deliver the intelligence is the support given while implementing the intelligence itself.

The opportunity is not simply to build AI.

The opportunity is to distribute intelligence.

Blog | Benjamin Armstead