The three economies of AI

Most of the AI cost conversation I hear is about models. Which model, how big, what it took to train. I understand why, because those are the numbers in the headlines. But if you run a business, I think that is the least useful place to be looking.

There are three economies running at the same time. Pulling them apart has been the most practical thing I have done for my own thinking this year.

Model economics is not your decision

The first economy is the model itself. Call it intellect, or cognition. This is where the enormous investment goes: frontier models, a handful of companies competing at the top end, capital at a scale most of us will never touch.

Unless building models is your business, this is not your decision. You are not going to influence what a frontier model costs to train, and you do not need to. The one thing worth knowing is that the generation behind the frontier is relatively cheap to run, and the supply of open weight models keeps growing. Past that, I would leave this economy to the people whose business it is.

Inference economics is the cost of effort

The second economy is inference, and this is where I see business owners either get a grip on the numbers or get surprised by them.

Inference is effort. When you run a model across work that is mundane, or repetitive, or well understood by the model already, you are buying effort by the unit. Tokens are that unit. I find that a useful translation, because effort is something every business already knows how to price. You have been buying it in salaries and contractor hours for years.

So the discipline is straightforward. If a piece of work takes a million tokens to produce, what does that cost you in real terms, and what is the alternative, which is usually human labour? Work that out before you scale, not after. The cost of one run tells you very little. The cost at volume tells you whether you have a business.

There is a related point about model choice. You do not need the bleeding edge frontier model for a mundane piece of work. Choosing appropriately for the task is the cheapest saving available to most teams, and it is the one they skip.

Judgment economics is where the scarcity moved

The third economy is judgment, and it is the one I keep coming back to.

The models produce a lot of high quality work now. That part is real. What has not gone away is the need for someone with experience to look at it properly. I have seen this in our own business. Work came to me that had been produced by people who do not have my years in the industry. At first glance it looked great. Once I applied the lens of experience to it, there were nuances that had been missed. These were judgment calls rather than obvious errors, which is exactly why the model did not catch them and a junior reviewer would not either.

That is what I mean by judgment: the crystallised intellect that comes from having done the work for a long time. Intellect is becoming abundant. Effort is getting cheaper. Judgment is still scarce, and I think it is becoming more important than it was.

The practical question is reach

If you accept the three economies, the question becomes how far your judgment can reach.

There is no use producing a thousand items of work if all of it still bottlenecks in the one person in your organisation who is the reviewer. The output has grown and the review capacity has not, so nothing has really been scaled.

This matters most in the technology channel, where the constraint has always been scarce technical expertise. AI can produce the work. Your experts still have to apply their expertise and put their name to it. So the question for a channel business is how to grow revenue and capacity at a rate your technical people can stay in the loop for.

If you have the experience, your value went up

One last thing, because I think a lot of experienced people are quietly worried about this. If you have years under your belt and you have been watching people move quickly with AI, I would look at it differently. Understand what you do, where you fit now, and where your value sits. In my view your value probably went up, provided you know how to position it.

AI is a multiplier. People say that and stop there. It multiplies in every direction, including producing a great deal of rubbish at speed. Used with some judgment, it multiplies the few people in your business who are hard to replace.

Watch the full conversation with Sidney Minassian on YouTube: https://youtu.be/hZuHCTjoxxY

Subscribe to The AI Take for one AI lesson a week, and what it changes for your business.

About 1KE:

1KE is the presales autopilot for technical sales teams. It turns discovery calls, emails, and technical documents into structured requirements and customer-ready solutions in minutes, so teams can respond while the deal is still warm.

Learn more → https://1ke.ai

Michael → linkedin.com/in/michaelchanter/ · https://x.com/1keceo

Sidney → linkedin.com/in/sidneyminassian/

Follow 1KE → linkedin.com/company/1ke-ai/

‍