It is 9 p.m. on a trading floor that never really empties. Somewhere a first-year analyst would once be three hours into a comparable-company analysis, squinting at filings, copy-pasting numbers that refuse to tie out. Tonight, that work is already done. A note was drafted, the comps pulled, the deck started - by software that was trained to think like a banker and never asks for a coffee. The software is called Rogo, and it has made itself useful to roughly 35,000 people who used to do that job by hand.
Finance runs on grunt work nobody wants to do
The dirty secret of high finance is that a great deal of it is data entry in a nice suit. Junior bankers spend days - sometimes nights that become days - pulling numbers from SEC filings, building the same valuation models, and assembling memos that say roughly what last quarter's memos said. The work is necessary. It is also slow, repetitive, and expensive, and it burns out the smart people hired to do it.
Gabriel Stengel knew this the way only someone who lived it can. He was an analyst at Lazard. His co-founder John Willett had done the same at JPMorgan. They had personally generated the all-nighters they would later try to abolish. The question that became Rogo was not "can AI write poetry?" It was narrower and more useful: can AI do the part of the job everyone hates, without getting the numbers wrong?
A senior thesis, a release of GPT-3, and a resignation letter
Rogo did not start in a garage. It started as a Princeton senior thesis - a chatbot for econometrics - shared by three classmates: Stengel, Willett, and Tumas Rackaitis. It might have stayed an academic curiosity. Then GPT-3 arrived in 2020 and made the founders suspect the timing had changed. So they did the unfashionable thing for people with offers from Lazard and JPMorgan: they quit.
The bet was specific. Not a general-purpose chatbot dressed in pinstripes, but a system trained explicitly for finance - one that understands market positioning, runs peer comparisons, and models valuations the way a deal team expects. In a regulated industry, "mostly right" is a failing grade. Rogo's wager was that a finance-native model, with citations you can audit, would clear a bar that consumer AI never could.
Gabriel Stengel
Ex-Lazard analyst. The one who sent the resignation email first.
John Willett
Ex-JPMorgan, Princeton CS. Knew exactly which workflows to automate.
Tumas Rackaitis
The third classmate. Turned the thesis into a model finance could trust.
Bloomberg's headline, April 2026: "Junior Bankers Sick of Grunt Work Build $2 Billion AI Tool to Do the Job." Sometimes the press writes your About page for you.
Meet Felix, the agent that works while you sleep
Most AI tools answer questions. Rogo decided that was not enough. Its flagship agent, Felix, executes - it screens deals, generates CIMs, handles buyer outreach, and runs data-room diligence on its own, across multiple steps, without a human re-prompting it at each turn. You can hand Felix a task by sending it an email, and it tailors what it produces to your role. The analyst and the managing director get different answers to the same question, which is exactly how a real team works.
Around Felix sits the rest of the platform: finance-trained search, comparable-company analysis, sector benchmarking, and document generation that drops straight into Excel, Word, and PowerPoint. It plugs into the systems firms already live in - SharePoint, CRMs, and data providers like Capital IQ and FactSet. The output is meant to be institutional-grade: models that tie out, memos with citations, decks you could put in front of a client.
Security first, magic second
Wall Street does not adopt software because it is clever. It adopts software because compliance signed off. Rogo built the boring part first: role-based access, audit trails, citations on every claim, and alignment with SOC 2, ISO 27001, GDPR, CCPA, and the EU AI Act. It runs inside each institution's secure data environment, built on Amazon Bedrock. The pitch to a chief risk officer is not "trust the AI." It is "here is the paper trail."
Four years, four rounds, one straight line up
The thesis becomes a company
Three Princeton classmates leave finance to build a finance-native AI after GPT-3 changes the math.
$7M to chase Wall Street
Seed funding to advance specialized generative AI for financial services.
$75M Series C · Sequoia leads
European expansion announced. Henry Kravis and Wells Fargo join the cap table. Valuation ~$750M.
Acquires Offset · partners with LSEG
Agents pushed deeper into workflows; trusted market data wired in via the London Stock Exchange Group.
$160M Series D · ~$2B valuation
Kleiner Perkins leads. Total raised passes $315M. 35,000+ users across 250+ firms.
The numbers a skeptic would ask for
Adoption is the only metric that survives a hype cycle. Rogo's is hard to wave away.
Valuation, by funding round
Three bars, one direction. The jump from January to April 2026 is not a typo - that is roughly a 2.6x re-rating in a single quarter.
Who is actually using it
The names on the list are the kind that usually build their own tools rather than buy.
Not a bot. A teammate.
Rogo's stated goal is modest in wording and ambitious in scope: to be every firm's most reliable, efficient teammate - expanding what a team can do rather than replacing it. The framing matters. "AI that takes your job" sells panic. "AI that does the part of your job you hate" sells seats. Rogo is betting on the second story, and 250 firms have so far agreed to hear it out.
The flagship agent is named Felix - Latin for "lucky" or "successful." You put it to work by sending it an email. The most powerful colleague on the floor doesn't have a desk.
The all-nighter, abolished
Return to that trading floor. The lights are still on - finance does not sleep, and neither, conveniently, does Rogo. But the work filling those late hours has changed shape. The analyst who once rebuilt the same model for the fifth time is reviewing one the agent drafted, and arguing about what it means instead of whether the cells line up. That is the quiet revolution here: not that the machine is smart, but that it is finally pointed at the work humans were never meant to enjoy.
Whether Rogo grows into the "operating system for finance" its investors are paying for is still an open question - $2 billion buys expectations, not certainty. But the thing it set out to kill, the analyst's all-nighter, is already on the endangered list. For an industry that runs on other people's time, that turns out to be worth quite a lot.