The Mind of Jean Jacques Janse van Rensburg: An API for Working With Me
Jean Jacques Janse van Rensburg is founder and CEO of SelfBuiltSystems, a frontier AI systems firm, and co-founder of . He built LNCELOT, an intelligence and prediction market platform, and Leyline AI. He produces the Jean Jacques Janse van Rensburg channel on AI, private equity, and venture capital, and runs business and AI strategy across a private equity portfolio, where he owns growth end to end: sales, marketing, and operations. His background spans venture capital, private equity, artificial intelligence, and behavioral economics.
I am an operator and investor, and I think about compounding systems: intelligence, capital, and organizations, all the way up to civilizations.
I came up not being the smartest kid in school. I grew up in the middle of nowhere in a small railroad town in South Africa called Volksrust. Remember the railroad. It matters later.
Before venture capital and private equity, I did a BCom in strategic management, and I dropped out of university in my second year. I went out of my way to learn from entrepreneurs who were actually really good at doing their thing. They told me: "Get out of your comfort zone. Pick up books on quantum physics, engineering, artificial intelligence (a new paper every week), investing, and economics."
I later did focused programs through Copenhagen Business School on applied neuromarketing. At 26, I started AI-focused studies through MIT. I now read as much as I can, and I treat every business problem as a physics problem: find the constraint, find the mechanism, remove what does not survive criticism.
This page is the API layer for working with me. Below are the interfaces: what I am interested in, how I think, and what I am building. Call them in any order.
I am interested in loops
The AI engineering field just converged on one word. At this year's AI Engineer World's Fair, the main stage belonged to loops: Ralph loops, loop engineering, loopcraft, software factories. Steinberger says stop prompting agents and start designing the loops that prompt them. Cherny at Anthropic says he writes loops and the loops do the work. Huntley's Ralph loop restarts an agent against the same spec with a fresh context window until the spec is satisfied, and the apparent waste is the point.
Strip the hype and there are at least four distinct architectures hiding behind the word: the execution loop inside a single agent, the task loop that restarts agents against a spec, the product loop that runs an entire codebase as a software factory, and the system loop where agents improve the system that improves the product. Above all of them sits a fifth loop that mostly goes unnamed: the oversight loop, where goals get set, budgets get allocated, and work gets culled.
Here is what I find striking. The field just reinvented my epistemology and shipped it as tooling. A Ralph loop is Popper running in production: bold conjecture, ruthless criticism, fresh context, repeat until the artifact survives. A software factory is an organization compiled into code. And the real lesson underneath the noise is the one I have built my career on: human judgment is migrating up the stack. The scarce skill is no longer doing the work. It is designing the system that does the work, and knowing exactly where a human must remain in the loop.
That top ring, the oversight loop, is where I live. It is where I have always lived. A portfolio operator is an oversight loop over companies. An investor is an oversight loop over capital. The tooling finally caught up to the org chart.
At SelfBuiltSystems we do not sell prompts. We install factories: the specs, the harnesses, the evals, and the checkpoints where your judgment stays wired in. Capability belongs to the machine. Agency stays with you.
I am interested in sovereignty
There is a second conversation running underneath the loops conversation, and it is the more consequential one.
Geoffrey Huntley made the argument earlier this year that open source was always a financial weapon by design: release something for free and you destroy the ability to make money from it. Linux was built and it broke Windows' grip on the server. The conjecture now is that the same weapon is being fired at nation scale, with frontier-class open models released for free while trillions are spent on closed labs. I hold that claim the way I hold every claim, as conjecture. But the question it forces is not conjectural at all: when your firm's operations run on intelligence you rent, what happens when the spigot gets repriced, rate-limited, or turned off?
Meanwhile the local AI builders keep publishing the other half of the argument: open models now trail the frontier by months, not years, and the gap keeps shrinking. And the sharpest insight from that camp is not about models at all. A model alone is not a system. What you are actually buying from a hosted provider is the infrastructure around the model: the search, the tools, the harness, the ingestion, the agents. That layer is what most firms are missing, and that layer is buildable.
So build it. Own the rails.
This is the work at SelfBuiltSystems. We operate across the full stack of intelligence, bits, atoms, and electrons. Intelligence is the model layer. Bits are the software and the harness around it. Atoms are the hardware it runs on. Electrons are the energy that feeds it. Most firms touch only the top layer, and they rent even that.
We design and deploy private AI infrastructure for founders and firms who refuse to run their business on someone else's terms: dedicated compute you control, open models you can audit, and the complete operating layer around them, wired into loops with your judgment at the top.
I grew up in a railroad town. Railways were the infrastructure layer of the last industrial transition: whoever laid the rails set the terms for everyone who shipped on them. Intelligence infrastructure is the rail network of this one. I am back to laying rails.
I am interested in frontier AI as an operating discipline
I founded SelfBuiltSystems, where we build AI systems for founders and firms, and where I run a standing intelligence practice that decompiles every major model release into a working operator playbook. The gap between what frontier models can do and what most firms actually deploy is the largest arbitrage in business today.
Closing that gap is the work.
I am interested in capital and the machinery around it
I work inside a private equity environment where I own growth across the portfolio and build the financial models that decide where capital goes. I practice as a business founder first, and as an investor second.
The best investors I have studied are all operators in disguise.
I am interested in what happens when beliefs carry a price
I built LNCELOT because prediction markets are the purest expression of accountability: if you claim to know something, stake something on it. Most opinions are free, which is exactly what they are worth.
I am also exploring a new build, PROJECT Cognitive Compass, with outside partners. More on this soon.
I am interested in high-performing organizations
Organizations are the fundamental unit that scales an individual, and the best ones run on explicit cultures, ruthless assessment, and proof over promise.
In today's language: the best organizations were always loop stacks. Explicit specs, tight feedback, relentless culling of what does not work. AI did not invent the software factory. It just made the factory legible enough to automate.
I co-founded to build exactly this kind of organization. The name stays sealed until launch.
I am interested in how knowledge grows
My epistemology comes from Popper and Deutsch: knowledge advances by bold conjecture and ruthless criticism, and good explanations are the ones that are hard to vary. I treat every strategy, model, and belief I hold as conjectural and improvable. This is not a philosophy hobby.
It is the operating system underneath everything above, and, as of this year, it is apparently the operating system underneath the entire AI engineering field. They call it a loop. I call it how knowledge has always grown.
I am interested in the human machine itself
I train six days a week, every week, and I treat physical capacity as the base layer of the stack. You cannot run frontier software on failing hardware.
PS: I think Hyrox is cute :)
Ultimately, I am interested in compounding at four scales: the individual, the firm, the portfolio, and the civilization. The pattern is the same at every scale: build the loop, wire in the criticism, keep judgment at the top, and own the infrastructure underneath.
There is more signal and more noise in the world than at any point in history. This ecosystem is where I separate the two.
Subscribe for free to get my essays delivered to your inbox. Everything here will always be free to read. If you choose to subscribe and send me money, I promise to allocate the funds to good use.
Work & systems
- SelfBuiltSystems: frontier AI systems and private AI infrastructure for founders and firms
- LNCELOT: intelligence and prediction markets
- Current project: · in soft launch, announcement pending
- Frontier Model Intelligence: the standing practice that decompiles every major model release
- YouTube: AI, private equity, and venture capital
