At Talent Land 2025 I talked about how artificial intelligence became the new brain of Wall Street. Not as a metaphor: there are funds that stopped deciding by hand years ago, and the way they operate now shapes how the market everyone invests in actually moves.
What changed
For almost a century, investing well was a matter of having better information than the next person and reading it faster. That's over: today the information is essentially the same for everyone and arrives at the same time.
What sets a fund apart is no longer what it knows, but what it can process and how fast it decides. And there a machine always wins.
Advanced algorithms process enormous amounts of information in seconds, predict market movements, optimize portfolios and adjust positions in real time. No human competes with that on speed; humans compete on something else, and I get to that at the end.
Three houses that did it first
Two Sigma
A fund that behaves like a technology company that happens to invest. Its raw material is data —much of it non-financial— and its product is models. The underlying thesis: if you find signal where nobody is looking, you have an edge until someone else finds it.
Numerai
The strangest and the most interesting. Instead of hiring data scientists, it invites thousands of strangers to compete by submitting models over encrypted data: nobody knows which assets they're predicting. Numerai aggregates the best models into a meta-model and pays based on performance. It's an investment fund built as an open tournament.
WorldQuant
Its bet is volume: systematically searching for thousands upon thousands of small signals —"alphas"— knowing each one is individually weak. The advantage isn't the brilliant discovery, it's the machinery capable of finding and combining many mediocre ones.
What AI does well here
Processing what no team can reach: filings, news, prices, satellite data, web traffic, all at once
Finding patterns nobody looked for, because it isn't biased by a prior theory
Optimizing portfolios, balancing risk and return across more variables than anyone can hold in their head
Executing without emotion: it feels no fear and no euphoria, the two most expensive ways to lose money
And what it does badly
This was half the talk, because it's the half nobody tells.
It learns from the past. A model trained on ten years of data knows how to operate in the world of those ten years. When the regime genuinely changes, history stops being a guide and the model keeps trusting it.
They look too much like each other. If many funds train on similar data and reach similar conclusions, they all sell at the same time. AI doesn't just predict the market: it starts becoming a cause of its movements.
It can't explain itself. A model that is right and can't say why is a regulatory problem, not just a philosophical one. When it fails, somebody will have to answer for the decision.
What I left with the room
That the work left for people isn't competing on speed, it's deciding where the system must stop. What level of risk is acceptable, which operations need a human signature, what to do when the model asks for something that makes no sense.
That isn't nostalgia for manual work: it's systems design. And it's exactly what applies to any AI agent making decisions inside a company today. The question isn't whether you let it run alone. It's where you put the gate.
Frequently asked questions
How is artificial intelligence used in investing?
To process large volumes of information in seconds, predict market movements, optimize portfolios according to risk profiles and execute trades without emotional bias. Funds like Two Sigma, Numerai and WorldQuant have operated this way for years.
What is Numerai?
An investment fund that works as an open tournament: thousands of data scientists submit models trained on encrypted data, without knowing which assets they're predicting, and the fund combines the best ones into a meta-model, paying based on performance.
What are the risks?
That models learn from the past and fail when the market regime changes; that many funds train on similar data and end up acting alike at the same time, amplifying moves; and that the decisions are hard to explain to a regulator.
Who gave this talk?
John Olven (Jonathan Olvera) at Talent Land México 2025, Guadalajara, April 2025.