AI as a mirror for testing

In the last period I have seen many discussions about how testing should be done. AI seems to have brought a lot of attention back to topics that, in my view, were always part of testing. Everywhere you look, people talk about clear intent, shared context, avoiding assumptions, exploring variations and identifying risks early. And although these conversations sound new, the ideas themselves are not new at all. They have always been essential, even if they were not always treated that way.

For many years the focus in our industry was mainly on automated test execution. We talked a lot about speed, coverage and the number of tests we could run. Useful topics, of course, but they never represented the full picture. The more difficult part of testing was always about understanding what we actually want to build. It was about intent, context, assumptions, risks, behaviour and consequences. And when these things were unclear, they often stayed hidden for quite some time.

AI changes that dynamic. When intent is not explicit, it becomes visible immediately. When context is missing, the output feels incomplete. When assumptions are weak, the answers become confident but wrong. And when risks are not considered, the blind spots appear in a very direct way. In that sense, AI acts like a mirror. It reflects exactly what we give it, and nothing more. It does not hide the gaps; it shows them.

This is why it may feel as if AI is changing testing, while in reality it is only exposing what was already happening beneath the surface. Testing was never only about executing checks. It was always about thinking, understanding and exploring ideas before any behaviour existed. AI does not replace that. It simply makes it harder to ignore when these things are missing.

So for me, AI does not introduce a new way of testing. It makes clearer what was already important, even if it was not always treated that way. Testing has always started with understanding what we want to achieve, and only then looking at how we check it. That part has not changed.

What AI does is show more quickly when something is missing. If intent is unclear, it becomes visible. If context is incomplete, the result feels incomplete. If assumptions are not discussed, they suddenly appear in the output. In that sense, AI helps us see the gaps earlier, but it does not replace the thinking we have to do ourselves.

Two small examples from my own work illustrate this. While testing the new AI-Assisted Model Creation (AAMC) feature in TestCompass, I gave the AI plain text and it generated a model. Whenever my intent or boundaries weren’t explicit, the model added transitions or behaviours I never mentioned. The missing assumptions became visible immediately.

Another example is when I give the AI an incomplete requirement and ask for test cases. It often comes back with missing rules such as required fields or payment validations. None of which I specified. Those rules show exactly where my own intent or context was incomplete.

The real challenge is to keep that focus. It is easy to fall back into a pattern where we mainly look at execution again, simply because the tools become faster or more impressive. But testing has never been only about running things. It has always been about understanding first, and acting after that. AI does not change that. It only makes it more visible again.

.