What if product requirements doubled as prompts?

Soon, anyone will be able to paste some notes about a feature and get a finished design with code, ready to publish.

We’re not there yet. There are real constraints, like existing platforms and existing codebases. I wanted to see how far we could get today.

What I did

I took a real requirements doc and worked it into three different designs, each based on something I knew about our users. Along the way I wrote down:

• what I changed, and why
• how the model reasoned through each change
• what still worked against our real codebase

What I learned

The designs weren’t the useful part. The useful part was finding out which parts of our requirements docs were ambiguous. The model would confidently pick an interpretation, and half the time it was one nobody on the team had considered.

A requirements doc that produces a good design is just a well-written requirements doc.