AI for Test Debt In 2026

Recently, we have gone through a UI revamp of a decade old application. The new UI looks refreshing, it feels faster and cleaner. But all such revamp work do not always end on a happy note. While developing the new UI, we traded-off a few things, especially "tests".

We deprioritized fixing locators of our 300+ playwright tests.

I mean, that makes sense considering we had tight deadlines, and application stability was the first priority.

Now the system is stable!

We need those 300+ tests working again in our pipeline. But first we need to fix the locators.

In 2026, this task feels interesting to me for two reasons.

• I'm not treating this as a cleanup project. I plan to leverage AI heavily.

• I want to understand how effective AI really is for this kind of work where the problem isn't writing new code, but repairing the tests.

I don't have a single solution yet. I have hypotheses. Over the next few weeks, I'll be testing them.