As intelligence becomes cheaper, execution becomes more valuable

For most of the last thirty years, the scarce resource in a software company was the ability to write software. Clients paid for it by the hour, and the price reflected how hard it was to find people who could do it well.

That is changing quickly. Language models now draft code, translate documents, summarise research, write first versions of marketing copy and answer routine customer questions. None of it is perfect. All of it is cheaper than it was two years ago, and it keeps getting cheaper.

What gets cheaper

If you list the capabilities that AI is making cheaper, the list reads like the service catalogue of a typical agency:

• Programming
• Content creation
• Research
• Translation
• Design
• Administration
• Customer support

A business whose main product is one of these is selling something whose price is falling. Doing it faster with AI helps for a while. It does not change the direction.

What does not

Other things do not get cheaper when a model gets better. A customer who trusts you. A channel that reaches buyers in Germany at a cost that works. A workshop that can make a handmade product to a consistent standard. Data that exists only because you ran the operation. Knowledge of how an industry actually works, including the parts nobody writes down. A partner who answers the phone.

These are slow to build and hard to copy. AI can make them more productive. It cannot replace them, because they are not information problems.

What this means for what we build

We used to be a company that sold engineering. We still have the engineering, and we use it every day. But we now choose work by asking a different question: does this give us a durable position in a real business, or are we being paid for something that is getting cheaper?

In practice that leads us towards businesses where software and AI make a physical or operational process work better: personalised products that need fitting and production, sales processes that need measurement and quoting, operations that need customer conversations handled reliably. The technology is essential. It is not the product.

Leverage, not a substitute

There is a tempting version of this argument that says AI will run the business. In our experience that is the wrong lesson. The businesses that benefit most are the ones that already have something real: customers, a product, an operation. AI multiplies what is there. It does not create it.

So we use AI to increase operational leverage, not as a substitute for having a real business. The sentence is simple. Living by it means saying no to many attractive ideas, and doing a lot of unglamorous operational work on the ones we keep.