This place used to be called Prototypeverything. It was the mantra I lived by. It stood for a simple belief: you learn more by making than by only thinking and debating. A rough sketch in someone's hands answers more questions than a week of meetings.

That idea has since become default. With AI in our toolkit, a prototype that used to take weeks now takes minutes. Prototyping used to be a luxury: something we'd do only if time and budget allowed. Now it's simply how design gets done.

That's mostly good news. Mostly. Now that building is almost free, we make a lot more than we used to, and we have to choose between all of it. Which of these is the right thing? More often than not, things get settled in a room full of people with opinions, none of whom will use it.

AI Agents are good at creating something believable: a layout that looks fine, copy that sounds sort of right, a flow that seems to tick the boxes. In the end the code runs, so it's tempting to ship it and fix it later. But believable isn't the same as good. An agent can generate an interface, but it has never been confused by one.

The work moves to what building alone can't settle: three things that rarely line up. How we speak, how the technology works, and how the people using it think. The misalignment shows up in flows that make sense to us but not to our users. In words that mean one thing to us and something else to our audience. Aligning all of this is translation work, which is why this place is now called Speak Human. It doesn't matter what language we use ourselves; we should make things that speak the language of the people using it.

The only way to know if it does is to check with real people. Put it in front of our audience and watch what they do. Ask the right questions and hear what they see. We are not our users, and our brains are terrible reality simulators. A model trained on all of our brains doesn't fix that.