The Assay · 29 July 2026

The Tool Was Never the Hard Part

The companies that build the best AI in the world have just spent more than $5bn telling you the AI is not the hard part.

In the space of a few days this spring, OpenAI and Anthropic each stood up a separate company whose whole job is to put engineers inside other businesses and rebuild how those businesses work. OpenAI’s is reported at around $4bn; Anthropic’s, alongside Goldman Sachs and Blackstone, at $1.5bn. Money spent not on a better model, which they already own, but on the people who make a model useful once it meets a real business.

Sit with what that admits. The model was never the hard part.

Why nothing changed when you tried it

Most owners I speak to have already had a go. A few people use a chatbot, someone tidied up a spreadsheet, perhaps a report gets drafted faster than it used to. And yet the business is the same shape it was. The work has not moved.

That is not a failure of effort, and it is not because you bought the wrong product. It is the difference between two things that look alike and are not: dropping AI into the business as it stands, and first making the business itself explicit — writing down how it actually runs — so that AI has something solid to take hold of.

The unglamorous name for the second one is systematising. Before anything is automated, someone has to set out how the business really works: how it decides, where the knowledge sits, what would stop if one particular person were not in on Monday. Most "AI projects" skip that part, because it is slow and it cannot be bought in a licence. So the chatbot sits on top of a business it does not understand, and it changes nothing that matters.

What most people buy instead

The market is happy to sell you the skippable version. A tool. A licence. A strategy deck with a roadmap and a slide about "the art of the possible." Each feels like progress — money has been spent, something has been installed, there is a login.

I am not going to tell you those things are worthless. A good tool is a good tool. But none of them touch the part that actually decides the outcome: the business stays exactly as structured as it was, and the knowledge that runs it still lives in the same few heads. You have made the work a little faster in places. You have not made the business any less dependent on the people who hold it together.

The part the billions just bought

Here is what the labs worked out, expensively, and then said out loud. Marc Nachmann of Goldman Sachs put it plainly: the technology alone "doesn’t change your workflows or how you operate." What changes things is the embedded human work — people who sit inside a business, learn how it actually runs, and rebuild the work around what the technology can now do. The industry has a name for them: forward-deployed engineers. $5.5bn is roughly what it costs to rent that layer at the top of the market.

The data scientist Vin Vashishta frames the same gap from the other side: there is a difference between knowing about something and being able to do it, and a model that can describe your business with great fluency still cannot run it. The describing is cheap now. The doing is where the value sits — and the doing needs someone who knows what good looks like. In your business. Not in general.

Manufacturing learned this order the hard way, and its most quotable engineer spelt it out. Elon Musk’s five-step discipline for building any production system puts automation last: question every requirement, delete every part you can, simplify what survives, speed up the cycle — and only then automate. He is candid about why the order matters, because he broke it himself: the Model 3 line was automated before the work had been questioned, and he ended up scrapping automation he had only just installed. His verdict afterwards — the excess automation was his own mistake, and humans are underrated. Automating a process nobody has first understood does not remove the confusion. It sets it in concrete.

That is the reframe. Systematising comes first; automation comes second and depends on it. AI is what makes the first part affordable at your size — the patient, structured write-up of how a business works that, a few years ago, only a large company could pay a consultancy to produce. The labs have just confirmed that this embedded layer is the real product. Their market is the top of the pyramid. The same work, at the scale of an owner-managed business, is yours to build — and for the first time it is within reach.

It is also, not by accident, the work that makes a business depend less on any one person. A business whose workings are written down and structured is one that keeps running when someone is away, or moves on, or simply cannot be reached on the one morning it matters. The tool is how you get there cheaply. It was never the point.

What I’m not claiming

Systematising a business is slow, and most of its cost is senior attention rather than software — the people who know how the business works are the people with the least free time to write it down. It is not a product; you cannot buy it in a box, and anyone selling it as a quick install is selling the skippable version again. And the money the labs are spending does not trickle down to you — their engineers are pointed at companies far larger than yours. What is true is narrower, and worth only what it is: the work they have just priced at the top of the market is now affordable lower down, if you are willing to do the unglamorous part. That is the claim. Not that it is easy.

In No. 3 — Before You Trust the AI With the Numbers, I argued the numbers can be made trustworthy. This is the larger version of the same point — the tool is the easy half, and the half that matters is the patient work of making your business understandable to something other than the people who already run it. If that is work you would rather do properly than quickly, future editions of this newsletter are the rest of the answer. And if you would rather start on the unglamorous part than read about it — you know where to find me.

The Assay is written by Rudi Prenzlin at Scalable Growth, a UK practice that gives owner-managed businesses a fractional CFO and the operational structure to run without every decision going through the owner.

← All notes Begin a conversation