Jan Szilagyi

A market built to argue with itself

Name a company after a George Soros theory and you are telling people exactly how you think. Reflexivity - the idea that what we believe about markets bends the markets themselves - is the wager Jan Szilagyi made when he stopped trading other people's billions and started writing software.

Reflexivity, the company, sells a kind of confidence that is rare in artificial intelligence: it promises not to make things up. Szilagyi calls it a "zero-hallucination, plug-and-play investment analysis platform." In an industry where a confident wrong answer can vaporize a position, that constraint is the entire product. The platform knows that Iowa temperatures move harvest yields, that interest rates tug at bank stocks, and it lets a human pull those threads on command. "A human user is the conductor," he says, of "a vast array of analytical powers."

He describes the thing he was solving for as the "IKEA of financial information." Every part is in the box. Assembling it into something useful, before a market moves, is the brutal part. Reflexivity is the missing instruction sheet - and the wrench.

Killing tasks, not jobs

Most AI pitches promise to replace the human. Szilagyi's pitch is the inverse, and it is more interesting for it. "AI - in investment management - is killing tasks, rather than jobs, and this will reward experienced users," he says. The analyst who knows which question to ask still wins. The machine just gets them to the answer faster.

That belief shapes the architecture. Reflexivity is built to be agentic without being reckless: it can "judge partial results and correct course if it judges it is veering too far from the original plan." A self-healing research assistant, in other words, that knows when it is lost. Szilagyi is blunt about why the guardrails matter. "In finance, you cannot afford to have a tool that would hallucinate."

Five blocks to a yes

The introduction came through a college roommate's father, who walked him into Duquesne Capital's offices to meet Stanley Druckenmiller. Szilagyi left the interview unsure how it had gone. The offer reached him within five blocks. He dropped out of Harvard, took the desk, and in his telling has not looked back since.

Harvard is worth pausing on. He holds the record for the fastest Economics PhD the university has handed out - 2.5 years - written under Ken Rogoff, with John Campbell and Andrei Shleifer in his orbit. Before that came Yale, which he remembers as an "academic all-you-can-eat buffet": European history, Italian literature, experimental physics, before mathematics and economics finally pinned him down. He grew up between two countries, a Slovenian mother and a Hungarian father, then left for school and kept going.

What followed was a tour through the upper floors of global macro. Duquesne under Druckenmiller from 2005. A partnership at Hawker Capital. Fortress Investment Group, where the global macro book reached $15 billion. Then co-CIO of global macro at Switzerland's Lombard Odier. Two decades of reading the same data that everyone else had and trying to see the relationships first.

Global macro is a particular kind of trade. You are not picking a stock so much as reading the weather of an entire economy - rates, currencies, commodities, the way a temperature swing in Iowa ripples into a harvest and then into a price. The job rewards people who can hold a hundred moving variables in their head and notice when two of them have quietly started moving together. Szilagyi did that by hand for the better part of his career. Reflexivity is, in a sense, an attempt to externalize the habit: to take the web of relationships a veteran trader carries around intuitively and make it queryable, auditable, and fast.

From the trading desk to the founder's seat

When your old bosses write the checks

There is a quiet endorsement buried in Reflexivity's funding history. When the $30M Series B closed in October 2024 - led by Greycroft and Interactive Brokers - the participating names read like a roll call of the people Szilagyi used to work for and alongside. Stanley Druckenmiller. Greg Coffey. General Catalyst and SoftBank LatAm rounding it out. Druckenmiller, in fact, provided the earliest financial backing for the company.

It is one thing to leave the hedge-fund world. It is another to have its most famous practitioners fund your bet that the tools they used were not good enough. Total funding now stands at $45.75 million.

The Rosetta Stone moment

Szilagyi has a name for the shift he thinks is underway. He calls it AI's "Rosetta Stone moment" - the point where "machines and humans can understand each other better than they were able to do before." For two generations of financial software, the burden ran one way: the human had to learn the machine's language, its query syntax, its menus, its quirks. Natural language flips that. "This is where I think it's going to really lower the barrier," he has said, "for advisors to use a whole host of analytical tools that they previously might have shied away from."

That framing matters because it widens the audience. Reflexivity was conceived for professional portfolio managers, the people who live inside Bloomberg terminals and risk dashboards. But Szilagyi's larger thesis is that the same engine can serve the advisor and, eventually, the individual investor. He likes the word "wealthcare" - wealth management reimagined as something closer to a basic service - and he is unsentimental about the size of the opportunity. "Demand for wealthcare is virtually limitless," he says. On CNBC he has gone further, arguing that AI tools will ultimately replace financial advisors outright, even as the platform he builds is designed to put more power in those same advisors' hands. The tension is the point: kill the task, keep the operator, and let the market decide how many operators it still needs.

The early machinery was less glamorous than the vision. Toggle AI ran on in-house machine learning and natural language processing, fed by financial market data from Refinitiv, packaged as a cloud application. Co-founded with Giuseppe Sette, the company set out to process vast data streams in real time and make the output intuitive - not just to specialists, but to anyone trying to interpret what a market was doing and why.

The house view

Better questions, faster answers

The through-line of Szilagyi's career is not the pedigree, though the pedigree is real. It is impatience with the gap between a question and an answer. A trader lives or dies by how quickly the right idea can be tested against the data. He spent twenty years closing that gap by hand. Now he is closing it for everyone else.

His own forecast is characteristically hedged and honest: "The evolution of AI is incredibly fast and exciting - but also unpredictable." He has appeared on CNBC arguing the gap keeps shrinking, and he is clear-eyed that the edge does not vanish, it migrates - to whoever asks the sharper question. That is a comforting thing for a former analyst to believe, and possibly a true one. Either way, he has built the machine that takes the bet seriously.