I often can’t put into words what I have in mind. I know it when I see it.
With people, that slows things down. With AI agents, it gets expensive, because they’re fast. Give an agent a vague brief and twenty minutes later there’s a finished feature. It works, the tests pass, and it isn’t what I meant. Now I either explain what’s wrong with something that already exists, or I start over.
So the order is fixed: pictures first, then code.
Before anything gets built properly, I want to see it. A mockup, a throwaway prototype, a first image of how it will look. A plan that exists only as text doesn’t count, however detailed it is. Text is where misunderstandings hide. A picture shows them.
A picture is quick to judge. I can point at the part that’s off: too loud, wrong place, that one. The next version is usually closer. When there’s a real choice, I’d rather see two options side by side than answer a list of questions about them.
The prototype is allowed to be thrown away. That’s what it’s for. It costs minutes, and it answers the expensive question early: is this what I want? Only once one version sits do we build it for real.
The same goes for the end. When something is done, I don’t want a report saying it works. I want to see the real thing, a screenshot or a short video of the product as it actually is. Green checks are nice. Looking is better.