Everyone is racing to build AI agents. I keep wondering if that's the wrong race.

Scroll the leaderboard right now and you can predict the next launch before you see it. It's an AI coworker. It lives in your Slack. It answers for you, drafts for you, books for you, follows up for you. The pitch is always the same: do more, automatically, so you don't have to.

I've been building long enough to feel the pull of that pitch. "Indispensable" is a comforting word. It sounds like a moat. It fundraises well.

But here's the pattern I keep noticing in my own life as a user.

The AI tools I tried once and forgot were the ones that did everything for me. Impressive in the demo. Slightly hollow by week two.

The ones I actually kept are the ones that made me a little sharper, then quietly got out of the way. I used them, I got what I came for, and I didn't need to open them again for a while. That "not needing it" was the sign it worked, not the sign it failed.

That's the part I think we get backwards as builders.

When your success metric is "how dependent is the user," your incentives slowly rotate away from solving their problem. A tool that genuinely cleared your plate would see you open it less. But a product measured by daily engagement learns the opposite lesson. Keep the loop open, manufacture a reason to come back, leave a little something unfinished. Not because anyone's evil. Because that's what the metric rewards, and metrics have a way of becoming the goal.

And I think people feel this even when they can't name it. It's why "trust" and "does this thing actually respect my time" are suddenly the words builders are using this year, instead of "look at this demo."

So here's the test I've started applying to everything I build and everything I use.

If it worked perfectly, would I use it more, or would I finally be done?

An agent is designed so the honest answer is "more, forever." A good tool is designed so the answer is "done, for now." Same model underneath. Completely different product.

I don't think one is objectively right. Plenty of real problems genuinely want an agent that just handles it. But the whole feed is betting on one answer, and I'm increasingly convinced the more interesting products are hiding in the other.

Curious where other builders land on this. When you imagine your product working perfectly, does usage go up forever, or does someone get what they needed and walk away? And which one were you actually optimizing for?