Are We Finally Learning Through AI How to Describe Requirements Properly?

If anything positive comes out of this whole AI craze, then please let it be that we’ve now finally learned how to describe requirements clearly.

In the past, you could say, “we absolutely need this feature now, and exactly like this, because otherwise the customer won’t sign!” Weeks later, you didn’t end up with what you wanted — you ended up with what you said.

Thanks to AI, this feedback loop no longer takes weeks, but maybe a few hours or days.

Because of the short time between requirement and result, we suddenly realize that it’s often not development that takes a wrong turn somewhere on the way to the solution: through their “work,” Claude and Codex mirror our input back to us. Anyone smart enough now takes another look at the input when the output is unexpected, and maybe realizes: “Ah, I see how that could have happened.”

I find this genuinely fascinating: through AI, we’re currently building systems everywhere to make the creation of software more straightforward and “unambiguous.” But the problems behind that haven’t existed only since 2024 — they have always been there. The difference is that communication and clarity have become tangible and traceable as the main problems.

What we’re learning in our dealings with Codex and Claude can be applied in every area of your organization:

• Ambiguity produces unpredictable results.
• Information overload crushes the one executing.
• Just because you can do something doesn’t mean you should, or have to.

Those who are experimenting heavily right now and realizing that AI coding agents (almost) only mirror our own behavior can set their organization up to be more productive and efficient. I hope many people have a lightbulb moment this year, and that we place far greater emphasis on the role of communication in product development.

And all it took for that was a chat window and a whole, whole lot of graphics cards.