Unlimited Leverage

Predictive models

The first point is a quick explanation of the state of the technology itself right now. Large language models are still, at their base, statistical sequence models. They are trained on large text and code corpora to model the next token given context. Modern assistants are then instruction-tuned, and often aligned with human feedback, but the underlying capability remains shaped by patterns in training data and by the prompt context.[1]

This does not mean they cannot abstract. It does mean that reliable abstraction away from the training distribution is the hard part: coming up with novel solutions to unsolved problems, new ways to solve problems, or new approaches in general.[2]

People already operating at the frontier of human knowledge in any given field, be it financial markets, science, especially the hard sciences, or art, are pushing human knowledge, looking for novel explanations, or finding new ways to express oneself. These things remain hard to outsource reliably to AI.

At the same time, what AI is very good at is doing things that it has been trained to do repeatedly.

The infrastructure barrier

In trading, this becomes concrete. A discretionary trader may want to formalize a strategy so that the fixed rules applied to markets manually are applied by an algorithm. A few years ago, this would only have been possible to a single person in very rare cases.

More often, it required a software engineer who was talented at what they do, somebody helping with data, and a whole team of people around the project. It is still important to code, or at least think like a software engineer. But today many such projects have opened up to individuals or small groups who do not have access to as much capital and workforce.

AI agents are very good at doing what has already been done, especially if it has been done several times and there are examples in their training data. This means they can help tremendously in building infrastructure, particularly when the work is well-scaffolded and familiar. The gains are not automatic. More importantly, they can get a project very quickly to the point where human thinking becomes relevant.[3]

This is the lowering of the entry barrier. Anyone operating at the frontier, or at least wishing to operate at the frontier, now has the leverage to get there quickly and efficiently, and focus on what is actually important for humans to do: coming up with novel explanations and ideas.

The edge

In algorithmic trading, the stack broadly needs three parts: data collection, backtesting, and execution. With the help of agents, it is possible to get there much more quickly, even alone. AI alone, without a capable operator, will not reliably produce the desired outcome. But it can get the project to the most important and differentiating point.

That point is coming up with the creative, novel ideas themselves. In algorithmic trading or market making, this means finding non-randomness in the data, inefficiencies in the data, and even more importantly, ways to exploit or smooth out these inefficiencies by trading them profitably.

The reason this will not easily be done by AI is the same reason AI is so good at building infrastructure. It is good at what it is trained on. But as soon as the real frontier is reached, AI should not be expected to do these things reliably and at high quality. Human beings are at a huge advantage here, because human brains are extraordinarily good at abstracting away from what has already been learned. Some people are better at this than others, but broadly speaking, human beings still have the advantage on this kind of abstraction compared with large language models. That is the relevant distinction.

At this point, it is not really about the quantity of work, which is much more the case when building infrastructure. It is much more about actually thinking, thinking deeply before committing to ideas, coming up with high-quality theories, testing these theories, iterating on them, thinking about how to improve the risk profile, how to construct the portfolio, and overall going through the process of alpha creation.

For this purpose, alpha inherently means having an edge: knowing something or doing something others do not know or do, or few know and do; seeing something in the data that only few see, while at the same time forming a falsifiable theory that can be tested, and that is non-random.[4]

This process is inherently an act of abstraction.

The most leveraged people

People operating at frontiers of human knowledge, but also entrepreneurs, scientists, and people who look for novel knowledge, for alpha, will be among the most leveraged people on the planet.

This is also the reason the doomsday scenarios around AI are incomplete. There is skepticism and there are questions about the unknown ahead. But the direction is already becoming visible, and it has had some years to manifest itself in society.

This can be a beautiful place, because it not only leverages the builders of society, people working at frontiers, but also empowers people who have not yet had access to the necessary workforce or the necessary capital to realize their ideas.

Access to workforce and capital is less determinative than it was. It is now possible to work together with Claude Code or Codex. Agent workflows are beginning to allow large tasks to be split across agents or subagents, with separate context windows, parallel work, and human oversight.[5]

Self-realization

This does not mean that AI and this revolution will only have winners. That would be the wrong argument. It does mean that access to value creation and the ability to realize ideas is being democratized. This single development is already a large revolution.[6]

If this power is learned and used well, more people than ever have a real shot at true self-realization. In return, this should greatly benefit society. If people are actually able to realize their own individual potential and ideas to a high degree and a high quality, this should lead to prosperous and net positive value creation in society.