engage
in the last post I said AI is an evolution of the interface. not a transformation of humanity, not AGI, just a better way to talk to the machine. here's what that looks like in practice.
the first thing I did when I got an actually good agent was say: let's build something that helps me work. not a todo app. something that matched how I actually think about work — what's in progress, what's blocked, what's next, what I'm just considering. it understood what I meant. the tool that came out fit how I think, not the other way around.
that's the inversion. you used to shape yourself to your tools. learn the software, adapt to its model, work around its limitations. now you describe how you think and ask for something that fits.
some of that is software. tracker-qt exists because I wanted the Haiku file manager on Linux and I described it precisely enough that it got built. mdread exists because tracker-qt needed it. five minutes defining requirements, success criteria — this part matters, probably more than anything else — and working software a few hours later.
but some of it isn't even software. the workstream thing is just prompts. a description of how I think about work, loaded into context. no code. the agent reads it and knows what I mean by "in progress" and what "next action" is supposed to look like. it's not an app. it's just describing yourself to something that understands you.
the old model was an app for every need, each with its own data model, its own interface, its own learning curve. you don't need Fantastical or Evernote.
you need a computer that knows you're supposed to get a quote for having the driveway repaired sometime this week.