The list of what not to build is the one that ships it

I shipped a working feature end-to-end in one session — unusual, and worth understanding why it happened so I can repeat it.

It wasn't speed or a clever shortcut. Before writing any code I wrote two lists: six things the feature would do, and thirteen things it explicitly would not. Auth. Hosting. Persistent history. Streaming polish. Model-selection UI. All named, all ruled out in advance.

That second list did the heavy lifting. Every "while I'm here, I should also…" impulse hit a decision I'd already made when I was calm, not mid-build when the scope was tempting to stretch. The exclusions weren't gaps to fill later; they were the boundary that made "done" reachable today.

A plan tells you what to build. An out-list tells you when to stop. I've come to treat the second as the more valuable of the two — especially for myself, because my failure mode is adding one more good idea, not too few.