I love that people can now build software without spending years learning the craft.
But an app that runs is not automatically an app you can trust with customers.
Understanding the generated code helps. Enough to spot obvious flaws and question the decisions behind it.
But what if you cannot read every line?
You can still define what the product should do—and build a test suite around those expectations.
For example:
• What happens when a payment succeeds but saving the order fails?
• Can one user access another user’s data?
• Does retrying an action create duplicates?
• What happens when an external service is unavailable?
Start with the requirements. Define normal behaviour, failure scenarios and expected outcomes. Then use AI to help implement the tests.
Simply asking it to “write tests for this code” risks testing what it already built rather than what you actually wanted.
A well-designed test suite becomes an executable specification of the behaviours that matter.
You can treat parts of the implementation as a black box while checking its inputs and outputs. The tests give you confidence in those scenarios—and help catch regressions when the code changes.
If you cannot design that suite yourself, this is a good reason to bring in an experienced test engineer or consultant.
Testing is an established discipline. But I see a growing need to make it accessible to founders building with AI who have no engineering team behind them.
As generating code becomes easier, knowing what to verify becomes more valuable.
