If You're Using AI to Generate Requirements or User Stories: You're Completely Missing the Point

The Bottleneck Of Requirements is NEVER Writing Them Down

You’ve seen those PMs who generate PRDs and User Stories using AI. They polish their prompts and fiddle with them until the requirements are perfect.

To everyone using AI to generate requirements or write User Stories, you're completely missing the point.

The big mistake you're making is that you believe the bottleneck for good requirements is the speed and accuracy with which you write them.

This is never the bottleneck.

Before the age of AI, I was pretty darn good at writing PRDs, requirements and User Stories. With my background in Software Testing and Business Analysis, I would polish my requirements until they were impeccable with faultless acceptance criteria and exhaustive BDD scenario’s.

When I presented these requirements to my team: they were completely silent.

And then, when we picked up the work, they still missed many things. Even in the rare case it was actually perfectly written down (which we could only determine afterward).

I completely changed my approach, and we started writing the requirements together. It took longer, but the requirements were better and there were far less mistakes. In fact, I received far less questions during development and even after features were live.

The Real Bottleneck of Requirements

The bottleneck is your team's understanding of the requirements and when you follow the AI-generated approach you're making their understanding worse.

Yes you will have requirements sooner, but that's not the point. You can even do waterfall, and have 'perfect' requirements before development begins.

The whole point of good requirements is that they are an act of collaboration. You write them down together and clarify details as necessary.

And you don't start with the solution, but with the problem. Translating the solution to requirements restricts agency and ownership, because delivering the solution becomes the goal and not solving a problem with the solution.

Jeff Patton had the best explanation for why requirements should be an act collaboration. Look at this picture, what do you see?

You see a forest with some tents and cabins. If you ask me, I'll have a far richer experience where tell you it was a lovely cold night with me sitting on a porch reading a book with the rain gently pattering on the tent.

That picture serves as a reminder of what I already know, and good requirements serve as such a reminder of what you already know.

If I’d give you this picture, and ask you to explain what it means, I’m robbing you of a far richer experience you could have by actually having been there. By using AI you're trying to short-cut the experiential dimension that's crucial for great requirements. As I explained in my previous piece, you’re only ensuring someone will remain stuck in Plato’s Cave of Requirements.

Please don't do that!

We’ve already failed at waterfall. There is no need to repeat the misery of waterfall through the application of AI.

Requirements that are thrown over the fence can never be good, even if they are perfectly written down.

Collaboration will always work better than throwing things over the fence.