Probably Purple

Someone's writing a brief. Typing it out. Not thinking about formats or systems. Just thinking out loud.

"Dark theme. Hero up top, big bold. Pricing, three tiers. Button, make it prominent, probably purple, or something vibrant. Oh, and testimonials. Like three or four cards."

That's a real person. You can hear them thinking. "Probably." "Or something." "Like three or four." Those aren't accidents. Those are someone still deciding.

The Form

That brief goes to an AI. The AI has a job: fill in a form. Usually a JSON, a rigid structure with named fields. Theme. Hero. Color. Testimonials count. Little blanks. Each one wants one answer.

Color: what's "probably purple, or something vibrant"? Not one answer, is it. So the AI picks what fits. Purple. Three or four? Pick one. Three. Done.

And the hedge? "Probably." "Or something vibrant." Where would they even go? There's no field for them. No slot for doubt. They fall off the side. Never make it into the form.

The Turn

But wait. Who killed it?

Not the AI. The AI did what the form asked. The form is the one with no room for "probably."

The JSON murdered the hedge.

Silently. No warning.

And it gets worse. The next thing reading that field? Often another AI. That second AI reads "purple." And to that AI, "purple" isn't a hedge anymore. It's a fact. Builds on it. Passes it on.

By the third hand, nobody knows "probably" ever existed. The person who wrote "probably purple" gets a purple button back. Looks like they chose. They didn't. The form chose. For them.

The Real Problem

Here's the thing that kills me. This isn't an accident. This is the whole industry's direction. Structured outputs. Function calling. JSON schema. Every AI shop is racing to force human thought into rigid shapes. That's the roadmap. Turn the mess into the structure. Optimize for the shape, not the thinking.

And you do this yourself. Every day. You prompt an AI: "make it warmer," "like three paragraphs, but shorter maybe." You're still deciding. The AI picks a shape and hands you back structure. The "maybe" died in between.

Most formats hold answers. They don't hold thinking. "Still deciding" doesn't fit. "Maybe" doesn't fit. "It depends" doesn't fit. None of the words we use when we're in the middle of thinking.

That's every form. Every day. Millions of times. Every time a human thinks out loud and a machine translates it into something rigid, the thought gets smaller. Every. Single. Time.

Small thoughts build small things.

The Door

Here's the distinction that matters. A translator interprets what you meant. An encoder preserves what you said.

Right now, the whole industry translates. Every pipeline guesses at meaning and hands you back a clean answer. That's why your hedge dies on the way in.

We need a new format. One that holds the thinking, not just the answer. We need to stop translating and start encoding.

Something better than what we have.

I have a solution. Wanna see it?