People who meet me through Froots are sometimes surprised to learn I didn't come up through a traditional ML lab. I came up through markets. Before I was writing about agent architecture, I was a trader, and then I co-founded and ran a retail proprietary trading firm. This is the short version of how I got from there to here, and why the path makes more sense than it looks.
Starting in markets
I spent about three years trading before I built anything that other people used. Trading teaches you a particular kind of discipline that doesn't show up in a CS degree: you are wrong constantly, the feedback is immediate and quantified, and the only thing that survives is a process you can repeat when you're tired and the screen is red. You learn to separate a good decision from a good outcome. You learn that edge decays the moment it's crowded.
That mindset turned out to be the most transferable thing I own.
Building Traddoo
In late 2022 I co-founded Traddoo, a retail prop-trading firm. The thesis was simple: there were a lot of skilled traders with no capital and a lot of capital with no good way to find skilled traders. A funding and evaluation product sits in the middle of that gap.
Running it taught me the parts of company-building that trading never could:
- Operations are the product. Customers don't experience your strategy; they experience your dashboard, your payouts, your support response time. The boring infrastructure is the brand.
- Trust compounds or it collapses — there's no middle. In any business handling other people's money or expectations, every small inconsistency is a withdrawal from an account you can't easily refill.
- Most of the work is removing friction you stopped noticing. The biggest wins were almost never new features. They were the three-click flows that should have been one.
The pivot
By 2025 I was spending more of my time automating my own operations than running them. I had agents reconciling data, drafting communications, watching dashboards, and doing the repetitive research that used to eat my mornings. At some point the obvious question landed: the tooling I was building to run a company by myself was more interesting, and more broadly useful, than the company it was running.
That's how Soshi Labs and then Froots happened. The same instinct that made me build a funding firm — find the gap between a capable person and the thing stopping them — pointed straight at AI agents. The gap now isn't capital. It's leverage. Most people can describe exactly what they want done; they just can't be in ten places at once doing it.
What carried over
- Process over vibes. A trading system you can't write down is just a feeling. An agent you can't observe is just a guess. I care obsessively about making agent behavior legible — which is the whole premise behind building agents you can actually watch work.
- Risk is a first-class citizen. In markets, position sizing keeps you alive. In agents, the equivalent is interruptibility and review gates — letting a system act, but never letting it act irreversibly without a human seeing it first.
- Edge is in the boring parts. The flashy demo isn't the moat. Reliable memory, clean browser automation, and predictable failure modes are.
Where I'm headed
I'm building toward something specific: a company that runs almost entirely on agents, with me as the person who decides what should happen rather than the person who does all the how. Traddoo taught me how much of a business is repeatable. Froots is my bet that most of the repeatable part can be delegated to software that you can trust because you can see it.
If that resonates, Froots is the tool I wish I'd had when I was running everything myself.
Dylan Worrall is the founder of Froots and Soshi Labs, building always-on autonomous AI agents. He's based in Steamboat Springs, Colorado.