Generate enough and the user eventually hits a wall: what am I supposed to do with all of this? More summaries, more drafts, more bullets, more recaps, more options to review. It looks like value right up until someone has to use it. That's the lesson building Earmark drove home for us - users don't actually want more AI text. They want less unfinished work, and those are not the same thing.

More output isn't more progress. A long summary can still leave the PM staring at a blank ticket. A detailed recap can still leave engineering short on context. A polished follow-up can still miss the real next step. A full page of AI-generated bullets can still create work for whoever has to sort, edit, route, and decide what actually matters. Volume just relocates the effort; it doesn't remove it.

So the goal was never to generate more - it's to shrink the pile of work left unfinished. The best output isn't the longest one; it's the one that lifts the next burden. The ticket engineering can act on. The follow-up that's ready to send. The decision log that keeps the debate from restarting. The update that gives leadership clarity. The insight brief that actually helps product decide. None of those have to be perfect. They have to be useful - and there's a wide gap between AI that writes and AI that moves the work. Writing produces text. Workflow produces motion. Nobody leaves a meeting hoping for a lot of content; they leave hoping they won't lose the next hour turning the conversation into everything the team needs.

The question was never how much the AI produced. It's how much unfinished work it removed.

That reframe made us more skeptical of any experience that celebrates volume. Ten generated outputs aren't worth much if none are close to usable, and a beautiful summary doesn't help if the real work still starts after it. The better measures are almost the opposite of word count: Did it shrink the blank page? Did it preserve the context? Did it produce the artifact in the right shape? Did it make the next step obvious? Did it let the human review instead of reconstruct? The best output, as one person described it, is the one that makes you feel like you're already halfway through the task.

Humans still decide, still edit, still bring the judgment. They just shouldn't have to dig through more AI text to find the actual work. Less content that creates cleanup, less output that looks impressive but doesn't move - and more artifacts that reduce what's left to do. Users don't want more words. They want fewer open loops, better artifacts, and less unfinished work.