I’d rather play with the plan than read it

Planning with an agent through a chat interface usually ends in a document: phases, file lists, open questions, and a request for a yes. I skim it and say yes. What each of us meant only shows up later, in the diff.

Now I ask for a few versions of the idea I can try. A prototype I can click through, a behavior I can change. Rough is fine. It only has to be real enough to react to.

Trying them changes how I decide. I notice something I hadn’t thought to ask about, combine parts of two versions, or want to try something else entirely. Often that is how I find out what I wanted in the first place.

Bret Victor’s Media for Thinking the Unthinkable gets at this: seeing a system respond as we change it helps us reason about it. We can inspect what happens instead of keeping the whole thing running in our heads.

Or take a plan for a busy day. Give the writing another half-hour and something else no longer fits, even though there is free time left. Move the print pickup and watch the afternoon open up. Adjusting these sliders could be my reply to the agent, with the consequences visible before I settle on the values.

In a purely textual chat, every small experiment is another message, wait, and explanation to interpret. I’d like the agent to give me something I can think with, where changing my mind is as easy as moving a control and seeing what happens.