All notes
AI & innovation / Max Ek

AI makes judgment more important.

When making a first version gets easier, choosing the right thing deserves more attention.

Editorial draft · Max Ek · Prepared

Building with AI has changed the questions I ask. A first version can be much closer than it used to be. The work that interests me is deciding whether that version should exist, what it should do and what would make it worth improving.

My experience is as an operator building with assistance, not as a software engineer replacing years of technical experience with a prompt. That distinction matters. Being able to make something run is only one part of making it useful.

There are at least three decisions before the tool choice. Who is experiencing the problem? What does the current workaround cost them? What would count as a meaningful improvement? If those are unclear, a prototype can create the feeling of progress without changing much.

I like small experiments because they make those questions testable. Choose one workflow, one user and one outcome. Make the first version narrow enough that you can understand what happened. Then decide what to keep, change or stop.

The temptation is to keep adding features because the next addition looks inexpensive. But each one still adds something to explain, maintain or rely on. The cost of producing a screen is not the whole cost of owning a system.

That is why I am interested in AI as leverage. A useful tool changes what one person or a small team can do. It can help carry context, explore alternatives or build a first working model. It does not remove the need to choose a direction.

I also want to separate a demonstration from a dependable workflow. If something worked once, I have evidence of that one attempt. I still need to understand failures, access, inputs and the human who will own it when the demonstration is over.

The more possibilities I can explore, the more I need a clear filter. My filter starts with a real problem, a bounded next step and an honest account of what the result proves.

What is one problem you understand well enough to test with AI this week?