Testing ... nowadays AI writes your tests. Who decides whether they matter?

For years, testing was a question of time. You knew what needed testing. You just never got around to it.

That problem is largely gone.

AI writes a test suite in seconds that used to take a team days. Unit tests, test data, edge cases, documentation. Ask for it and it appears.

But a new problem has taken its place.

From too few tests to too many that prove nothing

An AI assistant looks at your code and tests what that code does. Sounds reasonable. But it isn't.

Because if your code does the wrong thing, AI will write a neat test confirming the wrong thing happens correctly. Green check. False confidence. Slob.

We now see this on almost every project. A suite of hundreds of tests, a coverage number everyone is proud of, and it still breaks at go-live.

Why? Because nobody asked the question AI cannot ask: what was it supposed to do?

What AI does exceptionally well

Be honest about where the gains are. They are real.

• Volume. Generating test data, writing variants, expanding coverage. Tedious work that now costs almost nothing.

• Speed. A test suite for new code, immediately.

• Consistency. AI never forgets a null value, an empty list, a negative number.

• Translation. From acceptance criterion to concrete test case in seconds.

• Maintenance. Updating tests after a refactor is no longer the reason to postpone the refactor.

Anyone not using this is leaving money on the table.

What AI does not do, This is where our work begins.

• Knowing what is at stake. AI does not know that one field on that screen determines whether a donation is tax deductible. The consultant who was on site last year when it went wrong knows that.

• Knowing the unspoken requirement. Every client has rules written down nowhere. They live in people's heads, in habits, in "that's how we do things here." AI reads your requirements. Not your organisation.

• Sensing that something is off. An experienced tester looks at a screen and thinks: a user is going to fill this in wrong. That is not logic. That is a thousand hours of watching real people use real software.

• Prioritising. You can test everything. But what must never break? That is a risk judgement, not a technical question.

• Carrying responsibility. When it goes wrong at the client, a person sits at the table. Not a model.

The new division of labour

Roles are shifting. They are not disappearing.

AI does	People do
Write tests	Decide what needs testing
Generate edge cases	Decide which edge case carries real risk
Increase coverage	Decide whether coverage is the right goal
Explain failing tests	Decide whether the test is wrong or the code
Make it faster	Be accountable

Put simply: AI supplies the execution, people supply the judgement.

Four things you can do on Monday

1. Let AI write your tests, but not your acceptance criteria. Those come from the client, from a conversation, from someone who knows the domain. Feed them in as input, never take them as output.

2. Review AI tests on intent, not syntax. The code is almost always correct. The question is whether the test guards the right thing.

3. Put your experienced people at the front, not the end. They used to review last. Now they need to look first, because they define the brief.

4. Stop measuring on coverage. Measure incidents after go-live. That is the only number your client feels.

Why this is our position

At CRMSolver B.V. we do not build teams that AI replaces. We build teams that can direct AI.

Our certified talent from Amsterdam Zuidoost, Suriname and the Caribbean learns two things at once. The technique, and the judgement.

The first you can learn quickly.

The second you earn by working on live client projects with real consequences. We have more than 2 decades of experience working with a broad range of clients.

That is where our value sits.

Not in who generates a test the fastest.

In who knows which test protects the client.

Your partner that helps you grow while making a difference.