Lore

I wrote my first code on a TI-84 in sixth grade, automating my math tests with programs where I could plug in numbers and run to get the answers. I would write games during class to keep myself occupied. Then nothing happened for ten years. In late 2021, bored in a lecture hall, I wrote a Python script to automatically lowball people selling climbing gear online. Six months after that, I discovered machine learning, found the GPT-3 paper, and new ideas came faster than I could build them.

I spent the next two years exchanging long grind sessions on nights and weekends for more dopamine. I coded during class, after work, whenever I could find a gap. In 2022, I spent ten hours trying to configure nginx on an EC2 instance and gave up. In 2025, I deployed OpenClaw to Fly.io with Tailscale networking and a live filesystem mount to my laptop in two hours. One afternoon, I had an agent spin up a GPU cluster on Modal, build an inference pipeline, download a model, deploy it, write a CLI to talk to it, and run tests. One day I was waiting in the Ticketmaster queue for Tame Impala tickets and asked an AI how to bypass it. It refused. Within a day I had an uncensored model running on cloud GPUs, offering unbridled knowledge (and power). That same year, I became a full-time director of agents at work, proving that what worked for personal projects held up in production. None of this was possible two years earlier. The technology got increasingly better and I became increasingly leveraged.

The thesis is that systems compound exponentially. Software builds on software. AI accelerates the compounding. Build the right system and ride the exponential. And pray it's not a sigmoid.

The agentic tools I'd built for myself started shipping production code at my day job. Then the reverse happened. Work taught me how to organize: mission, strategy, OKRs, projects, tickets. I started running my personal projects the same way.

Every project I worked on needed the same scaffolding: project management, agent configuration, deployment infrastructure. Building each one refined the scaffolding for the next, and eventually the projects started cross-pollinating. The Linear workflow I built for one project became the operating system for all of them. The custom MCP server I wrote to talk to Linear got used by every agent. The docs site I built to document one system became the reference for the whole company. The safe CLI, which let multiple agents work in parallel without stepping on each other. Pre-commit hooks steered code quality without human review. Then a full code review system replaced the human review entirely.

On the product side, the same compounding happened: formalizing projects, triaging tickets regularly, keeping docs current, strategic planning sessions. BENCORP crystallized around all of it. It was a joke until it wasn't. Every project feeds the system, the system expands what I can attempt, and the leverage compounds.

BENCORP is designed to build itself. Tools produce better tools, agents refine the systems that run them, and a growing knowledge layer means less gets lost between sessions. The goal is a system where maintenance is automatic and the only interesting decision is what to point it at next.

The goal isn't to make money. It's to reverse entropy. Leave the world in a higher energy state than it was a day ago. Making music is reversing entropy. Solving an unsolved problem. Building something physical. Every BENCORP venture has to pass the test: does this create more than it consumes?

The long-term targets are hard problems. Biotech, energy, robotics, healthcare. Problems where the data is messy and the tools are bad. The personal software is a training ground for systems that will eventually point at something bigger.