Letting Go of Control: Testing, Certainty, and the Future of Software

What keeps me curious about generative AI isn’t just the models or the capabilities — it’s the mindset shift it demands from us. We are moving from a world built on determinism to one shaped by probability. From code that behaves exactly as written, to systems that offer a spectrum of responses — none exactly the same, but all valid in their own way. Back when I was a junior engineer, we used to crack jokes that computers were the best coworkers — they never talked back, never took coffee breaks, and always did exactly what you told them to. Turns out, that era of obedient machines is over — now our “code” has opinions.

In corporate life, certainty is king. We don’t just prefer it — we worship it. We build dashboards and roadmaps and KPIs around it. We mistake predictability for control, and control for competence. There’s a quote from Bezos on the wall at Amazon HQ: “If you know it’s going to succeed, then it’s not an experiment.” And yet, so many leaders call for innovation while clinging tightly to control — demanding bold thinking from their teams, but denying them the space to take real risks. They want the fruit of experimentation without the mess that comes with it. The truth is, innovation isn’t born from control; it’s born from risk, from trying and failing, from stepping into the unknown without a guarantee.

And yet, so much of leadership today is an illusion of certainty. The idea that we can foresee and plan every outcome. But as GenAI enters the scene, it challenges this illusion. These systems don’t operate like the code we’re used to. They aren’t guaranteed to give the same answer twice. They are probabilistic by nature, fluid, and often surprising. Prompt Engineering exposes how poorly we communicate as leaders and how language can be intrepeted in many different ways.

This changes how we build — but more importantly, it changes how we test. We can’t rely on the old logic of “if it works once, it should work every time.” That logic assumes a static world. But generative systems live in motion. They learn, they adapt, they respond to nuance. And so must we.

Testing in a GenAI world, then, becomes less about binary outcomes and more about boundaries. Less about proving something is always true, and more about understanding the range of what might be true. We’ll need to embrace variation, not eliminate it. We’ll need to build systems that are resilient, not just correct.

And perhaps most radically, we’ll need to accept that not all errors are failures. Some are simply signs that our systems are behaving more like people — imperfect, unpredictable, but capable of surprising beauty.

In this new era, we don’t just test software. We make peace with the unknown. And in doing so, we create something more beautiful and alive.