I'll say the provocative part plainly: I'm building toward a company that runs almost entirely on AI agents, with me as the only human. Not because I dislike teams — I've built one — but because I want to find the real answer to a question that used to be hypothetical: how much of a company can one person operate if the repeatable work is delegated to software you can actually trust?
This isn't about replacing people
The "zero-employee company" framing gets misread as anti-labor. It's the opposite of how I think about it. As a solo founder, the alternative to an agent doing a task isn't an employee doing it — it's the task not getting done, or me doing it at 1am instead of the thing only I can do. Agents don't take a seat from someone. They take the backlog of work that never justified a hire and would otherwise rot.
Running Traddoo taught me how much of a business is genuinely repeatable: reconciliation, reporting, first-pass support, research, the same five-step flow run a thousand times. That work is real, it's necessary, and almost none of it requires me specifically. It requires someone reliable. That's the gap.
What makes it possible now
Three things had to be true, and they finally are:
- Agents can act, not just answer. They drive browsers, edit files, call APIs, and run commands. Output became action.
- They can remember. Memory that survives restarts turns a clever assistant into something that accumulates context about your business over weeks instead of forgetting you every session.
- They can be trusted, because they can be watched. This is the one people skip, and it's the one that matters most. Delegation requires trust, and trust requires legibility — which is exactly why Froots shows every action in a timeline you can pause.
Without the third, the first two are a liability. An agent that acts and remembers but can't be inspected is just a faster way to make a mistake at scale.
The founder's job changes shape
In this model, my job isn't to do less. It's to do different. I spend my time on three things agents can't own:
- Taste and direction — deciding what should exist and what "good" means.
- Judgment at the gates — reviewing the irreversible actions agents queue up before they happen.
- The relationships — the handful of human conversations that actually move things.
Everything between those is delegable. The leverage isn't that agents are smarter than people. It's that one person plus a fleet of legible, persistent agents can hold far more surface area than one person alone — without the coordination cost of a headcount.
Why I'm betting the company on it
I'm not theorizing about this from the outside. Froots is the tool I'm building and the way I run the company building it. Every capability I ship — the browser layer, the memory system, the interruptible timeline — gets stress-tested against my own operations first. The dogfooding isn't a marketing line; it's the development process.
The honest status is that we're early, and "almost entirely" is carrying real weight in that sentence. But the direction is clear, and every month the line between "I have to do this" and "an agent can do this, and I can verify it did" moves further in one direction.
If you want to run yourself at that kind of leverage, that's the whole point of Froots.
Dylan Worrall is the founder of Froots and Soshi Labs, building always-on autonomous AI agents. He's based in Steamboat Springs, Colorado.