The cost of AI has changed. The question has not.

When DeepSeek released a model competitive with the best available tools at a fraction of the expected infrastructure cost, the reaction in most business circles was surprise. In AI circles, the reaction was closer to confirmation of something already suspected: that the cost of capability was going to fall faster than the market had priced in.

That shift has real consequences for Norwegian SMEs. The price objection that stopped many small businesses from exploring AI in 2023 and 2024 is largely gone. Running capable AI systems no longer requires enterprise budgets. The tools available to a fifteen-person company in Frøya today are meaningfully similar to what a large corporation in Oslo can access.

That is the change. The question it raises is the same one it has always been: what problem are we actually trying to solve?

Cheaper tools do not make unclear problems clearer. They just make it cheaper to build the wrong solution.

I see this pattern regularly. A business owner reads about AI, decides it is time to do something, and starts looking at tools. The tool search happens before the problem definition. The result is a solution in search of an application - and usually a shallow implementation that does something impressive-looking and not much else.

The Clarify step is not optional even when the technology is affordable. Particularly when it is affordable, because affordability removes the forcing function that expensive projects impose. When something costs a lot, you define the problem carefully before spending. When it costs less, that discipline evaporates and you end up building faster toward something you have not properly defined.

The businesses that are getting real returns from AI right now are not the ones that moved fastest. They are the ones that asked the clearest questions. What specific task takes how much time and produces what kind of output? What does the right answer look like, and how would we know if the system got it wrong? Where does human judgment need to remain in the loop, and where is it just a legacy habit?

Those questions are operational, not technical. They do not require expertise in AI. They require someone willing to look at the actual work and describe it honestly.

In Trøndelag, the businesses I find hardest to help are not the ones who are sceptical about AI. The sceptics ask good questions. The hardest are the ones who are enthusiastic but vague - who want to "use AI" as an outcome rather than as a means to a specific operational improvement.

Clarity first. Tools second. Always.

Murphy Alex builds operational AI systems for Norwegian SMEs from Frøya, Trøndelag. IPRESTANDA is at iprestanda.com.