Time for a Pivot (Sort Of)
Not changing what the factory is, just what I do there!
Two weeks ago I got back from the most relaxing trip of my life in Norway and was thrust into a room on the UC Berkeley campus with 5,000 of the smartest people in AI, all excitedly chattering about the new thing they're working on. Half the people in that room are trying to build an epic, category-defining infrastructure company — myself included! But here's what I took away:
The hard problems I care about are still hard problems. The good news is that there are dozens of people smarter than I working on them. The bad news is that even I have a hard time cutting through the noise to tell the shape of what we're all building apart.
More importantly, no one has cracked a fully-autonomous factory that pumps out high-quality products end-to-end. The innumerable AI slop factories humming right along don't count, and I refuse to become one of them. I take some comfort in that: the missing last mile isn't just my failure. It's the state of the art.
Meanwhile, I'm sitting on a shelf of impressive 95%-complete products that I use regularly but haven't seen the light of day. People aren't getting the benefit of using them, and I'm not getting the signal (or the revenue!) they'd generate out in the wild.
I did this to myself: for months I treated my intervention in their end products as almost cheating. If my agents couldn't get to 100% on their own, it didn't count. That rule kept the factory honest, but it also quietly built a backlog.
So, about the sort-of pivot: I'm not changing the shape of the company — I'm temporarily changing what I do there. I've been playing mastermind, designing the factory floor and the agents themselves. For the next two months I'll assume the agents can get each product to the point where it's up to me to put a bow on it and get it out the door and into your hands. That's solopreneurship for now: the architect also gets to mop the floors and clean the toilets. Shipping will provide much more interesting data for the factory itself than another month of tuning it.
One of my mentors said it best: if nothing is coming off the conveyor belt, it's not a factory — it's a warehouse. And it's much less sexy to say I run a warehouse with a bunch of AI agents running around inside of it. I've stopped counting the full end-to-end miss as failure; it's just not solved yet. But watching all this useful work collect dust while working on it? That's failure.
That's also the ideal collaborator I'm most hoping to find: someone to serve as a forcing function, who helps put on the last bits of polish and never more than is needed. If that sounds like you, get in touch. Meanwhile, the first box is out the factory door today: Naggler, a clean, agentic attention ledger for those among us who do too much. (There is an adorable hedgehog!)