We test, test, and test again. But still get issues. AI to the rescue.

blog@dws.team
October 7, 2025
2 months ago
We test, test, and test again. But still get issues. AI to the rescue.

From web applications to spaceflight, what we produce is so fantastically complex, one tiny thing can bring it all down. We test, but how to cover all angles?

I'm a big fan of space technology and because I am from New Zealand it’s only natural that I root for NZ company RocketLab.

In a recent article in the Wall Street Journal I read that the companies’ new rocket Neutron would launch for the first time at the end of the year, placing itself as a direct competitor to SpaceX.

I hope nothing bad happens during that maiden flight.

Software and hardware systems in spaceflight are rigorously tested, as you can imagine. But some still fail, and if that happens during launch, it's often catastrophic.

RocketLab has had its share of hardship, but no failure was as jarring as the explosion of the Ariane 5 on June 4 1996.

That day was the maiden flight of the Ariane 5, a spacecraft built by the European Space Agency. 37 Seconds into the mission, the rocket exploded. Analysis showed that the craft lost balance, which triggered it to self-destruct. The root cause was determined to be a software bug that caused an overflow, one of the most basic issues imaginable.

Us more down-to-earth software folks see overflow errors quite frequently. Simple analogy: it’s like trying to fit a square peg in a round hole.

Ariane’s issue was a little more complicated, you could say that the peg only went square at a certain speed.

That it wasn’t discovered during tests of the Ariane is equally simple. Due to time constraints and human bias, the particular case that caused the error was not tested.

Ariane 5 used software that was developed for Ariane 4. It wasn't tested again because “it had always worked”. But the 5 went faster than the 4, and the peg went square at that faster speed. With disastrous outcome.

Space companies like RocketLab learn from such mistakes. But there’s only so many hours in a working week. That's where AI comes in.

With AI, far more scenarios can be tested. For instance, in “hardware-in-the-loop” systems, where the combination of hardware and software can be tested without the need of costly and sometimes dangerous physical tests, AI helps by creating scenarios at a speed previously unattainable, with more edge cases than humans could imagine.

But you don’t need to run a rocket company to have AI help you. In fact, we were building tests using AI before we were using it to generate business logic.

The software we build isn't as complex as a spacecraft by a long shot. But very complex nevertheless, with thousands of minute details, where the failure of any one could bring the whole application down. Just as that single, minute software bug caused the Ariane 5 to explode.

Luckily, our “disasters” are much less consequential. Downtime, lost revenue maybe, user frustration, it doesn’t compare. But we want to make our customers and our users happy. Which is why we test, test, test. And we are glad that we now have AI to help us.