Using AI, Part 3#

Lately I have been thinking again about the continuing evolution of AI, its use by businesses, including my own, and the limitations, restrictions, and risks that are starting to become more visible as the industry matures.

For a while, a lot of AI usage felt like a flat-rate service. You paid a subscription and got access to a very powerful capability. But that model now appears to be shifting. Usage-based pricing, API limits, model tiers, and higher compute costs are all starting to reveal the real financial cost of using AI at scale.

Some companies that once looked like very good value may become less competitive as pricing models change. The industry seems to be searching for a new equilibrium between what providers can afford to offer and what customers are willing to pay. In other words, less competition based purely on cost, and less subsidising of the true compute cost behind the scenes.

There is also the added issue of regulation and government intervention. Recently, Anthropic found itself having to restrict access to some of its most advanced models for foreign nationals, including those outside the US, due to national security concerns. In a way, it feels like they may have hyped too close to the sun: when you repeatedly present your models as world-changing, economically disruptive, and strategically important, you should not be surprised when governments start treating access to them as a national-security issue.

This presents a real problem for businesses that build themselves around a pure AI play.

If your product depends entirely on access to one specific model, what happens if that model becomes unavailable in your country? What happens if the price changes dramatically? What happens if the provider alters its terms, usage limits, safety rules, or API behaviour?

In that situation, you may suddenly find that the business you thought you had is much more fragile than expected.

It is another reminder of the old lesson: do not keep all your eggs in one basket.

Businesses should now add risks such as “lost access to AI services” and “AI costs becoming too high” to their list of operational risks. That does not mean AI should be avoided. It means it should be used with some thought.

For me, the sensible way to use AI is as an accelerator, not as the foundation of the business itself.

AI is excellent for research, development, idea refinement, writing assistance, technical exploration, and general productivity. I use it heavily, and I get a lot of value from it. But it does not run my business. I do.

That distinction matters.

I think about AI in a similar way to how I think about conventional cloud computing. You are using someone else’s compute resources to enhance what you are doing. There is nothing wrong with that, as long as you understand the costs, the dependencies, and the risks involved.

The important question is not “should I use AI?”

The better question is:

“What happens if I lose access to this AI service tomorrow?”

If the answer is “it would be inconvenient,” that is probably manageable.

If the answer is “my business stops working,” then you may have built too much dependency into the wrong part of the system.

AI is powerful, useful, and worth using. But like any external service, it should be used with a clear understanding of what it is, what it costs, and what risks come with relying on it.

I run Protoworks (NI) in Northern Ireland, helping small businesses design prototypes and build custom software. If you have a hardware or software idea you’d like to sketch out, feel free to get in touch.