I've been building bespoke agent systems since ChatGPT came down the pipe and changed the world. I've built a lot of them. So many different kinds.
It's the best job, and I love doing it.
But lately, watching the flood of OpenClaw clones and me-too agent apps hit the timeline, I've been thinking about purpose.
Building for the sake of building is never bad. Tinkering is how we get the accidental innovations that power the world.
I stand by that.
But once you start talking about real products, and production, and what deserves to exist in the marketplace, purpose matters more.
Does the world need another agent app now that every developer and their mother has taken a swing at one? Sure. If it has a reason to exist.
But this is the part I keep getting stuck on.
If we already have tool-capable agents everywhere, why do we still have more agents than tools?
That feels wrong to me.
Like the smell of a missed opportunity.
Most people who buy into AI tooling already have inference time on frontier models, and a stack they're happy with.
We're generating more code than we ever have in all of human history.
So why not make more and better tools?
Software that works with the models people already trust.
Legible, beautifully animated tooling. Cheap tricks on live canvases. Sticky interactions that feel more like entertainment magic than traditional software, while somehow still getting serious work done.
Stupid, silly, delightful features that users love, but enterprise buyers would never think to ask for.
Then let those users infect their employers with it, spreading your tooling deep inside the corporate beast.
Sell them once per product cycle.
Share your code. Build a community around something people use because they love it, not because you trapped them inside it.
Save your money.
You can vibe code anything.
But devotion is still an asset that pays dividends.