A startup valued at $2 billion is moving through the financial world with a focused mission: build the vertical infrastructure that will take over traditional banking work and position itself as the Bloomberg of the AI era, right as artificial intelligence spreads through every layer of Wall Street.
While the broader market chases the grand narrative of general-purpose models from OpenAI and Anthropic, this company has chosen an intensely vertical and deep path, from helping investment banks create presentations and Excel models, to managing entire M&A transactions end to end, and now expanding into wealth management.
In a recent interview on the well-known investing podcast Invest Like The Best, host and Positive Sum CEO Patrick O'Shaughnessy sat down with Rogo co-founder Gabe Stengel for an extensive conversation, breaking down the company's business logic, its subscription model, and his ultimate vision for the future of Wall Street.
The Core Competitive Logic: Why OpenAI and Anthropic Cant Win in Finance
The biggest question in the market remains: if large models keep getting smarter, why doesnt Wall Street just use ChatGPT or Claude directly?
Stengel's answer is that the financial industry requires more than intelligence; it demands extreme compliance, auditability, and deep integration with the underlying business operations.
"The thing you build has to be perpendicular to their direction," he said. "On the surface you might also do a chatbot, but underneath there's a huge amount of stuff they will never build."
His core thesis is that financial services are made up of countless niche markets, each with different data sets, different definitions of what "good" looks like, and different regulatory requirements. "Just going deep into those workflows, those data sets, you can build a $5 billion revenue business. For Anthropic, that's like stopping to pick up a penny on the side of the road. Their path is from $100 billion to $1 trillion."
Stengel offered two concrete examples of the "dirty work" that creates the moat. First is the MNPI compliance pipeline. "If you're handling market-moving trades, the compliance requirements are extremely precise. Every single action has to be auditable, and if regulators come in to review, you need pixel-perfect pipeline engineering," he explained. Second is the data room. "If a large public company acquires another, both sides need to exchange data securely. You need a compliant, secure data room, and it can't just be a static Dropbox folder. It has to connect to your agent workflows. OpenAI and Anthropic will never want to build a data room business."
"I don't think OpenAI or Anthropic will ever want to build a data room business," Stengel said. "If you want to be the exchange for all of high-end finance and capital markets, you need to own the intelligence, but you also need to own the venue, the communication place, the workflows, and all the data that goes into it."
This last-mile infrastructure is precisely where Rogo builds its defensible position.
Business Model and Performance Vision: The Bloomberg of the AI Era
On commercialization and pricing, Rogo currently uses a per-seat model similar to Bloomberg, with its customer base concentrated among large investment banks, not only because they have the most seats, but because they serve as the distribution channel for both talent and data across the entire financial world.
Stengel makes no secret of his admiration for the Bloomberg model and laid out Rogo's future development path. First, use a little AI and data to get in the door, much like the early Bloomberg terminal. Second, build out complete analytics and workflows, evolving from a copilot to full autopilot, where AI can write investment committee memos or due diligence questionnaires with 100% accuracy. Third, provide communication channels between counterparties, not for humans to trade, but for AI agents to negotiate and transact directly with each other.
On future guidance and pricing innovation, Stengel proposed a highly ambitious outcome-based pricing model: "For me, if I could skip token-based and usage-based pricing, what if I just charge you for every great investment idea I give you? Or charge per quarterly report that's perfectly delivered to your LPs? As an investment bank, if I can produce a complete confidential information memorandum and charge based on that, I'd rather get to that state."
He revealed that the product roadmap is only about 1% complete, with 99% of capital markets innovation still ahead.
Stengel also pointed to private markets as Rogo's core battleground, precisely because the plumbing there hasn't been built yet. "All the coordination, standardization, information mining, and actual trading in private markets is done manually. In public markets, so much is already automated."
Extreme Productivity Explosion: 90% of Enterprise Value Shifts to Software
The impact of AI on Wall Street productivity is already immediate and measurable. Stengel mentioned a managing director who hadn't touched Excel in 25 years and can now directly use AI to produce work on his own. "What used to take three days of back-and-forth with an analyst to produce a five-page client document, he now does himself in 10 minutes."
Stengel believes the financial industry is experiencing unprecedented structural disruption. "Over the past 10 to 20 years, being a capital allocator or an investor was a great business. There were natural barriers and inertia, especially in private markets. You raise one fund, then another, and it's hard to screw it up. But this is the first time that every investment institution and every bank is saying: 'I have to completely rethink my approach, and there are going to be hundreds of AI-native investment firms and AI-native banks attacking my business model very soon.' Private equity, hedge funds, investment banks haven't faced a true innovator's dilemma in a very long time."
He delivered a stark warning for traditional financial institutions: enterprise value is shifting. "If 90% of your enterprise value today is in people, your best investors and bankers who bring in the deals and revenue, then in 10 years, the world's best investment companies and banks will have 90% of their enterprise value not in people, but in software, data, and systems."
For the buy side in public markets, Stengel painted a vivid picture: "Imagine if every portfolio manager at a hedge fund had 10,000 AI agents communicating with each other, discussing ideas, reading notes, and debating around the clock. Then after 24 hours of relentless debate, they present you with just one great idea."
For private markets, he predicted that capital raising would become as simple as retail investors buying stocks on Robinhood: "I think KKR will be able to figure out in 5 minutes whether it can sell a portfolio company to another PE firm, not 5 months. Asset pricing time will compress dramatically."
Internal Company Brain: Non-AI Users Get the Dunce Cap
To achieve this AI-native transformation, Rogo's internal management is extremely aggressive. The company operates a corporate brain called "Shrek" where all internal conversations are recorded and fed into a shared memory bank and sales enablement tool. More dramatically, Rogo places intense emphasis on AI tool adoption rates among employees.
"Every month I get a report showing how many AI tools each person in the company uses. The bottom performer in each department gets a picture of a dunce cap printed and posted around the office," Stengel said. While he acknowledges it has a playful quality, it's designed to create a powerful incentive structure that forces everyone to integrate AI into their work instincts.
As Stengel put it, the innovator's dilemma in finance is just beginning: "There is no part of the financial business that you consider sacred and immune to AI."
Chapter 1: Introduction
Gabe Stengel: What we're seeing now is that these models are smarter than anyone I know, smarter than anyone around me. Imagine if every portfolio manager at a hedge fund had 10,000 agents, and these agents spent all day communicating with each other, discussing ideas, reading reports, and debating. Then after, say, 24 hours of uninterrupted debate, they bring you just one investment idea. Ten years from now, the world's top investment firms and best banks will have 90% of their enterprise value not from talent, but from software, data, and systems. What do you start doing? You start figuring out how to take everything that's accumulated in the minds of your best people and inject it into a system that you own and operate. No part of the enterprise is sacred. No part is exempt from being completely overturned.
Patrick O'Shaughnessy: We've talked about this fundamental question a few times in recent years, so let me start there. You're essentially trying to build a superintelligence tool for investing, helping investors get faster, better, cheaper, and easier. But increasingly it feels like we're already replacing core functions, even the work of very smart analysts and portfolio managers from just a few years ago. How do you see this trajectory, both from what you've experienced and observed, and where it's heading over the next two years?
Gabe Stengel: I think predicting two years is actually easier to extrapolate than predicting 10 or 20 years. Because in the next two years, the best investors will be trying to figure out how to reshape their firms and themselves. If you look at what happened in market making and quantitative trading, Jane Street took 15 years to build its dominance. The best investors in the world today will spend the next two to five years figuring out how to integrate AI into their work. You know, Dario has a famous quote that everyone will have a data center full of geniuses, or a nation's worth of genius in a data center. Imagine if Goldman Sachs, Millennium, or Citadel suddenly had a nation's worth of genius. What would they do? They'd probably need a considerable amount of time to figure out how to change the way they work, how to leverage it, how to integrate it into their systems. I think applying AI to the full investment life cycle is the biggest challenge facing every good investor in the next five years.
Patrick O'Shaughnessy: There are already many companies on this trajectory. Cognition is very famous, and their ads now literally say: "Remember Devin? Now it's really good." So many companies that are now clearly excellent with great products had phases in their AI business where the product was actually terrible, and now it's become outstanding.
Chapter 2: Building Rogo
Patrick O'Shaughnessy: If you think about the capability jumps we've seen from the underlying models themselves, can you tell the story of Rogo and its product in the same way? Define the phases yourself, however you want to frame it, and what it could do at each stage up to today.
Gabe Stengel: Yeah, actually before we formally started, I tried to create Rogo twice and failed both times. In high school, a friend whose father was an investment banker wanted an app to track the stock exchange ratio between two public companies during a merger. We tried using very old AI technology and it was a disaster. Then in college, before GPT-3 was released, we published a paper on AI-assisted econometrics and financial econometrics, and tried to commercialize it, but it completely didn't work.
The real start came after GPT-3 was released and before ChatGPT appeared. So in Rogo's early days, you could feel how magical it was. You could demo cool things, but nothing actually worked. I mean, to the extent that things started to really work, those phases are closely tied to model iterations. o1 Pro, and then around Opus 4.5. o1 Pro was the first version that gave you enough reliability to at least be a decent search tool. You could say, help me calculate some financial metric for this company over the past 12 quarters, and it would reliably do it without frustrating you enough that you'd rather do it yourself.
By Opus 4.5, around late last year or early this year, these models could basically handle any task a junior investment professional or junior banker could do, as long as you gave them the right instructions and context. I think for many applied AI companies, there's a "first mover disadvantage." You think you know where the world is going, so you want to build for that end state, but the models aren't there yet. So people try it and say, this is terrible, this is garbage. We went through that. But what we see now is: if you were right about the end state, if you knew where the models were heading and kept building toward that, then when the models actually arrive, it's magical.
For me, the best product validation is the feedback we get every day. People say constantly: this is changing how I work, I save hundreds of hours every month, I can now do things I never could before. So I get smarter and make better decisions. The experience is delightful and I love using it myself. It brings joy to my daily work because the user experience, attention to detail, and polish are clearly tailored for me and people like me. That's the most beautiful part of product building.
Patrick O'Shaughnessy: Would you attribute that to taking seriously all the compliance, regulation, workflow, and last-mile integration, and then the models matured enough that it all suddenly became extremely valuable?
Gabe Stengel: Beyond that, there are details about understanding how people in these institutions work and building specifically for that. Let me give an example. We made it extremely easy for a bank's managing director to send presentation edits via email, which is how they typically work anyway. Before, they'd send edits to an analyst. Now they send the same edits to our AI analyst and get a revised version back in 20 minutes instead of two days. Meanwhile, the system notifies the junior analyst working on the deal, shows them what's happening, and provides a full audit trail of every small change so they can step in if needed. This entire user experience and process allows financial professionals who might not have logged into a computer in a decade but know how to annotate on an iPad to truly adopt and use AI. These details of building a product that's truly right for a specific end user are things you only know if you have a spider-sense for the job.
Chapter 3: 10,000 AI Agents
Patrick O'Shaughnessy: Of the most cutting-edge things you can do, which impress you the most? Like, what kind of work can it handle?
Gabe Stengel: Oh, I think the coolest thing we're doing is applying innovations like something we call Multibook. Imagine if every portfolio manager at a hedge fund had 10,000 agents, communicating all day, bouncing ideas off each other, reading research reports, diving deep. Then after a full 24 hours of uninterrupted "debate," it presents you with just one idea. The reason this works is that investors will pay $50,000 for one truly great idea, and in almost any other field, it's hard to justify consuming that much compute just to get a single insight.
What we're seeing now is that these models are smarter than anyone I know or interact with. The key is the "plumbing" — connecting it to your context, telling it your investment thesis, telling it how you work, and integrating it into your daily routine. So building all that plumbing, actually gathering the data and context, is what I consider the true frontier.
Patrick O'Shaughnessy: One thing I've always found interesting about a product like this is that you have your own ideas about what it should be used for, but users can use it in creative ways. So you can also learn what people actually want to use it for. If I take a "God's eye view" — it's Monday morning in New York, and lots of Rogo users are probably using it enthusiastically right now. If I could see every active instance of this product, what would I see? Who are the users? What are the main use cases? How much variance is there? Tell us about how it's being used now.
Gabe Stengel: Although in many ways I think public market equities are the best use case for AI because all the data is publicly available, so you just need to be as smart as possible — that's a very, how should I put it, worst-case framing. Our early users and core market are actually people I'd call "deal makers" — people doing deals, buying companies, selling companies, or helping push transactions across the finish line.
So a lot of what we do is both making people smarter and actually getting deals done. How do you prepare a data room? How do you break down a data room? How do you coordinate calls with third parties to discuss data room content? How do you go through all the initial steps of a deal from launch to closing? If you looked down from above at all Rogo users, they're either the seller or buyer on a deal, using it to prepare all the thinking and materials to help drive execution — whether that's populating the data room with things like the business model, client description decks, answers to due diligence questionnaires about customer concentration; or on the other side, all those agents poring over data and mapping it against the institution's investment thesis to determine whether this concentration risk is below what our fund can accept.
Users access the system through all the classic channels — email, chatbots, proactive alerts, that kind of thing. But Rogo is already deep inside many of these institutions' backend systems, because when you're moving a deal forward, it's not just the people involved who need to know what's happening. You need to update your CRM, your portfolio monitoring systems, and how you distribute information to LPs afterward. So half of our business scope is actually below the waterline of the iceberg — interacting with these different system records based on what humans do throughout the deal process.
Patrick O'Shaughnessy: So you're serving deal makers now. When do you think you'll be able to give the same answer about serving junior analysts at public market hedge funds? Their workflow is a different process, though a defined one. You're right that my first instinct is also that public markets are the best place for this because there's so much data available. When do you think that shift happens?
Gabe Stengel: I think for our business, we need to have all the domain knowledge required to be a good public market investor. And I don't know what that takes — I've never done it. In fact, I haven't spent nearly enough time with people who are truly great at it. We need to recruit domain experts, and then based on that, figure out how to apply the systems we've already built to that market. I have a very strong belief that the underlying systems, tools, and infrastructure we've built will be extremely valuable in that market. But right now we need a good "chef" who can put it all together and create that final-form product and last-mile delivery for public market investors.
Our board has actually been pushing me to consider expanding our target customer base beyond the core investment banking business. But the reality is, in the "deal maker" vertical, there's been an incredibly rich depth and market size. And my ultimate vision — truly becoming the complete infrastructure for private markets, enabling people to transact very efficiently — matters much more to deal makers. In public markets, that infrastructure and those exchanges already exist. So I do want to serve them because I want to serve the best and smartest users who have vast knowledge of the companies and industries they track, and I want to make them smarter, which sounds very cool. But just by focusing on the space I'm in today, I can build a massive company.
Patrick O'Shaughnessy: Interesting. So a conclusion from this is: a lot of the opportunity in building vertical AI businesses is in places where the infrastructure hasn't been built yet.
Gabe Stengel: Yes.
Patrick O'Shaughnessy: And then put your...
Gabe Stengel: Right,
Patrick O'Shaughnessy: on top of that.
Gabe Stengel: Exactly. I mean, part of what makes private markets so attractive is that everything is done manually — the coordination, standardization, due diligence, and actual trading. Public equity markets, by contrast, have a large portion already automated.
Chapter 4: Skills That Still Matter
Patrick O'Shaughnessy: Based on what you know, which skills do you think investment professionals — broadly speaking, public or private markets — should consider becoming more valuable over time? Which become less valuable is relatively obvious. But let me flip the question: what are we supposed to do? AI can now do due diligence, investment committee memos, dissent handling, and all that kind of work end to end. That's a big part of what junior people in investing do. So what skills do you think will still be really important a few years from now?
Gabe Stengel: Let me start by saying I've worked in finance for two years, so I'm as much a student of this field as anyone, and I'm deeply interested in figuring out how to use the technology. But you know, the world's best investors have an ability to figure out what really matters and apply judgment across a range of topics. I think whether AI can eventually replace that is unsettled. When the "move 37" moment happens, when Lisa Doll sees something I could never see, if that starts happening in public equity markets, that will truly change what matters. If it really changes how everything operates, I think the core skill belongs to those who can go out and gather data and information that others don't have and feed it into the model. If you can spend time in the field, if you can talk to experts, if you can build a network of relationships that informs your model, maybe you don't need to personally compute your own "move 37" to make an excellent public equity investment, but you can actually feed your model with data no one else has.
Patrick O'Shaughnessy: Maybe now is a good time to talk about what I imagine as the "pipeline." For example, if Rogo outputs something very useful, tell me the entire system that produces that result. I want to understand every component of that system. What I'm most interested in is the data you own, use, buy, and build, and how you handle data, how you think about data, and how you think about models. On one hand, I'm curious about your specific business. On the other, I think a Rogo-like company will emerge in basically every vertical. So I'm curious what can be abstracted and replicated into other professional services or other interesting verticals that end up being disrupted by AI advances. Please go into detail: what exactly are the data and model pieces, and how have they evolved over time?
Gabe Stengel: Early on, Rogo was like a Rube Goldberg machine with about 60 different model calls. A question comes in, you first need to determine: which companies is Patrick asking about? What are their tickers? How do I feed those into API calls for Bloomberg, FAX, or some internal dataset? After the data comes back, I need to call another model to stitch it all together. As models get smarter, you want to be less prescriptive, less Rube Goldberg, and think about what's the simplest, best, highest-quality tool. It's like, if the world's smartest person showed up tomorrow and tried to be an excellent banker and investor, what tools would they need that don't just rely on raw intelligence? What are the data tools? What's the way to gather information? What's the way to review their own work? And how do you present results and push them back into the systems that need them?
So we've spent a lot of time thinking about what all the data inputs are that a good banker or investor really needs to do their job, and then we've spent a lot of time thinking about all the compliance and regulatory requirements when doing that work, to ensure that if one day a Delaware judge includes AI inputs and research as discoverable evidence, you've actually done everything the right way, ensuring information isn't mixed improperly. Because the reality of AI investment judgment and AI banker output is: you'll be able to see the full provenance chain of how these outputs were created. We've also spent a lot of time benchmarking these models and creating different types of evaluations and datasets, so we can always determine which model performs best, which has the lowest cost in token terms, which has the lowest latency, and route tasks to the most appropriate model type.
Chapter 5: Beating OpenAI and Anthropic
Patrick O'Shaughnessy: I'm sure the most commonly asked question you get is: how do you eventually compete with Anthropic and OpenAI? They both view finance as an important vertical, and it's one of the few areas where you can see them publicly talking about it and taking it seriously. In the long term, what do you think you can do to create a differentiated position against what they can or are willing to do?
Gabe Stengel: I mean, the thing you build should be perpendicular to what they want to build. Sometimes you might also do a chatbot because it helps you get to market faster. But you have to be clear that underneath the surface, there's a massive amount that those large labs will never build, and that's what we need to build for finance or any other vertical. When you think about financial services and capital markets, how many businesses are there where going deep into these workflows, these datasets, and these ways of working, can create more than $5 billion or $10 billion in value? On top of those extremely complex, highly specific problems, there's enormous market space and spending. The whole financial industry is really a collection of different niches, each with different datasets, different definitions of "good," and different regulatory requirements. We can build a $5 billion revenue business just by going deep into these areas and building the systems of record to manage them. For Anthropic, it's like stopping to pick up a penny on the way — because their path is from $100 billion in revenue to $1 trillion. There's tremendous depth behind these systems, and what truly needs to be built goes far beyond just "intelligence."
Patrick O'Shaughnessy: Is there a favorite example of the kind of "last mile" hassle you had to work through?
Gabe Stengel: Yes, let me give you one. Imagine you're dealing with market-moving trades or M&A, and you need to ingest MNPI. Just in terms of auditability, you have a set of compliance requirements: what can be tagged, how to connect it to the investment firm's or bank's internal systems so that if an auditor or regulator ever asks to see your records, all that "plumbing" is perfectly in place. That's one example. Another is, if you actually want to get deals done — say a large public company is acquiring another large public company and data needs to flow back and forth — you actually need some kind of data room, a compliant, secure, coordinated place for data exchange. Ideally, it's not just a static Dropbox folder, but a system deeply integrated with how you work, with your agents, with your workflows. I don't think OpenAI or Anthropic will ever want to build a data room business. And if you truly want to be the exchange for all of high-end finance and capital markets, you need to own the intelligence, but you also need to own the venue, the communication place, the workflows, and all the data inputs that flow in.
Patrick O'Shaughnessy: That's interesting. I was originally going to ask about data and models, but it sounds like the "harness," or infrastructure, might be the most important of the three.
Gabe Stengel: Think about when Claude Code came out, and the fundamental difference between Claude's agentic mode and OpenAI's ChatGPT. The models were pretty close at that point, but Claude's harness and presentation layer were much better, allowing the model to fully leverage its long-horizon capabilities. That's why Claude saw an explosion in usage and significant expansion. It tells you that how you harness these models matters enormously. I think people underestimate the nature of "intelligence." Humans can be highly autonomous and accomplish a lot, not just because we have strong memory, high IQ, and knowledge, but because there are all sorts of microservices in the brain: how do you store knowledge? How do you retrieve it? How do you trigger it? Emotions are a way to trigger these different microservices. That's also why a great investor often has excellent judgment — they have a "spider-sense" for certain phenomena in the market, and when they see that pattern, it automatically triggers a memory of some historical event that supports a creative decision. All these small microservices are things that need to be built.
Patrick O'Shaughnessy: If I forced you to become an investor with the sole goal of investing in companies like Rogo — vertical application companies that I'd think of as "plugging AI capabilities into an industry" — based on your experience, what traits would excite you most? Whether it's the industry itself, or the founder/builder and their methodology?
Gabe Stengel: A few things. First, the industry itself needs enough complexity and depth — in data types, in the types of systems of record people use, in the types of deployment models — that you can spend a lot of effort solving these problems, forming a "wedge" that allows you to solve everything else. Because if any person can just walk into this industry and sell a basic version of ChatGPT, Cowork, or Copilot without solving those weird integration problems, you won't have enough time to build things that are perpendicular to your core direction. So the industry needs to be close enough to the core market but with high enough barriers. That's the first point.
Second, I'd look for domain expertise in the team and founders. I didn't have unique domain expertise myself, but I had enough to get started, and I'm genuinely curious about finance, money, and capital markets. I grew up in New York, surrounded by people whose minds were full of high-end finance. I'm fascinated by it and want to understand it deeply. Then we built a team equally passionate about it. We now have over a hundred employees who previously worked at top global investment banks or investment firms, so we can continuously take each new model release and harness it for finance. Our job is essentially to capture every model change and figure out how to apply it in these institutions.
Third, and the last important trait, I'd look for a company willing to continuously reshape its core product, willing to tear it down completely and rebuild. I think if a company's product delivery method — like a terminal, or some very specific user experience or interface — can't be rapidly "self-disrupted," it's not agile enough to continuously reinvent itself every six months when step-changes happen.
Patrick O'Shaughnessy: Have you done complete teardown-and-rebuild examples?
Gabe Stengel: The story that most inspires me is Max Levchin talking about how Affirm rebuilds its core ledger technology every year. They do it for several reasons: one, it's the most interesting engineering problem, so all engineers want to work on it, and it's a great way to retain talent and get engineers deeply familiar with the company's core business. Two, it ensures the system doesn't fossilize but keeps improving. We do exactly the same with our harness and core agent systems. We're constantly examining it, realizing we haven't even reached a local optimum — in fact, with models changing so fast, reaching a local optimum is impossible. We need to continuously redo the whole system.
Patrick O'Shaughnessy: What can't models do yet that, if they could, would truly change the nature of the product?
Gabe Stengel: Compaction. That is, if you've had 100 conversations with an agent, how does it ensure it remembers the right things and compresses memory into the right token size that can be invoked every time, while maintaining enough coherence and context about who you are and what you care about, so it feels like you're talking to someone who knows you, progressively better? That's hard enough on its own, and when you think of agents not just as one-to-one relationships, it becomes exponentially more complex. Almost all agents today are one-to-one — you use ChatGPT alone, Copilot alone, Gemini alone. But once these agents actually collaborate with many colleagues, or work in a Slack channel with 100 people, operating across an entire company, the compaction problem becomes exponentially harder — because it's simultaneously talking to 100 different people and needs to coordinate across all those conversations.
So how do you take all that memory, all those interactions, and actually integrate it into the agent's "mental model" or "brain" so it's persistent and maintains real context across a series of interactions? That's where models still fall short.
Patrick O'Shaughnessy: How do you think about the main categories of AI software companies? We've seen companies like Cognition and Cursor growing at staggering rates. I'm particularly interested in you comparing this to the old software company taxonomy. I'm also curious how Rogo sells, and which category you'd put it in — usage-based? per-seat? something else? Can you talk about how people want to buy these products and what new business models are forming among AI software companies?
Gabe Stengel: We currently price per seat because our buyers are used to paying per seat. They categorize us in the same bucket as Bloomberg, FactSet, Capital IQ, and Pitchbook. So we have to build a very "people-intensive" business — every contract requires account managers, solution architects, sales engineers visiting, shaking hands, explaining how the product works and integrates. You can't sign one contract and have usage automatically grow 100x.
I look at how easy it is for Anthropic to sell to us — very easy. And the amount we pay them grows exponentially, with no human intervention in between, because it's a token consumption model. But many industries can't just ride the "token consumption tailwind," and enterprise sales is one of them, including ours. So it's actually quite interesting — we have to build a go-to-market machine at five times the speed of most enterprise sales organizations.
So I think there is indeed a category: the classic enterprise sales model, but with AI models as a tailwind to build a product that's 100x better. Then there's the other category: token intermediaries, token consumption businesses sold within enterprises to departments already used to usage-based pricing, like Cursor, Claude Code, etc. That way you can go further with fewer people.
What's your prediction for whether and how this changes in finance? I think every company needs to go through two pricing revolutions: first to some usage-based model, then to some outcome-based model. For me, if I could find a way to skip token metering and usage metering entirely, simplifying everything for the user, until I can say: "Patrick, what if I charged you every time I gave you a great investment idea? Or per quarter, I perfectly produce your LP report? Or as a banker, you pay per CIM you create?" — I'd rather get to that state than think about how to spread costs across tokens. That way of thinking doesn't produce agreement — if you spend $100,000 on tokens, you ask "did I really get $100,000 of value?" Nobody knows. But you know what a great investment idea is worth to you because you can see how much you actually made. And you know what a CIM is worth because you know how much advisory fees you charge these companies to get the deal done.
Patrick O'Shaughnessy: Per-seat pricing obviously can't adjust dynamically, at least not for the same customer. So how do you deal with the fact that this is, in some ways, misaligned with your customers' interests — if you do your job too well, they use more, your costs go up, and they become "worse customers"?
Gabe Stengel: The reality is, we've only completed 1% of our product roadmap, and 99% of capital markets innovation lies ahead. So what really matters is: we're a good partner, a good steward of their AI strategy, and they're willing to continue working with us in the future.
Chapter 6: The Bloomberg of the AI Era
Patrick O'Shaughnessy: It's fascinating to think about where this goes. You said you're only 1% done with the roadmap. From a product perspective, where do you see all this heading? Not the entire industry, but if you have that much roadmap left, can you describe it for us?
Gabe Stengel: I mean, think about how much of capital markets workflows, investment workflows, investment banking work is still completely manual, human-mediated. It's similar to other parts of financial services — 15, 20 years ago, everyone getting a mortgage had to go to a local bank branch and meet with a loan officer face to face. It felt very "human": I'm buying a house, I'm taking a loan, this is important, I need to talk to someone. No one thought this process could happen without human involvement. And now, 40% to 50% of mortgages are done directly through platforms like Rocket Mortgage online.
I think there will be a massive amount of innovation in how companies transact, how they raise capital, how they take on debt. I believe that in the next 10 or 20 years, it will become far easier for business owners or company employees to go online, click a button, and try to raise capital — just like someone can log into Robinhood today, click a button, and buy a stock. I think KKR will need 5 minutes to figure out whether it can sell a portfolio company to another institution, not 5 months. I think the time required for asset pricing will shrink by an order of magnitude. As a result, markets will become more transparent, more liquid, more efficient, and overall transaction activity will increase dramatically.
Patrick O'Shaughnessy: So I want to focus on a specific future for automated risk pricing — to use an imprecise shorthand. Say it's three years from now and everything you just said has come true. I can raise a few million dollars of debt or equity like filling out an online form, and the system prices risk and gives me a quote directly. Kind of like Opendoor, but for everything. To achieve this capital markets future, what do you still need to build that doesn't exist today?
Gabe Stengel: It's actually very similar to Bloomberg's strategy. I'll skip a lot of detail, but the outline goes like this: First, provide a little data as a wedge. Build on top of that the analytics and workflows everyone needs. Then provide an exchange and communications platform that lets users actually transact across asset classes that have been quite opaque until now.
Bloomberg has Bloomberg Messenger. For us, it's: first use a little AI as a wedge, then build out the full workflow, evolving from a copilot chatbot to fully autonomous tools. That way I can ensure 100% accuracy on your investment committee memo or due diligence questionnaire, then provide communication channels between counterparties where agents that can automatically do the work actually complete transactions for you. The difference between me and Bloomberg Messenger is: I don't need to build communication channels for humans to trade. I need to build communication channels for agents to trade across firms and investment institutions.
So if you think about what infrastructure truly needs to be built: imagine what kind of system would allow a large private equity firm to confidently let an agent negotiate deals on its behalf, communicate with all the third-party advisors on the deal — the quality assurance people, legal counsel, and so on — and then actually run an auction process where a group of institutions submits offers. The amount of software that needs to be built is enormous. I think sometimes we describe this as: for many non-standardized asset classes, it looks like an exchange.
Patrick O'Shaughnessy: Tell us about your current customer mix — how much is large banks, how much is investment firms, the ratio of public to private companies...
Gabe Stengel: Currently it's mostly large banks. The reason is simple: that's my background. I did a few years of investment banking focused on buy-side M&A, which was very interesting. So we targeted banks early, for a few reasons. First, investment banks are in some sense the distribution channel for the entire financial industry. Many people who later become great investors started as two-year analysts in Goldman's TMT group or something like that. Second, they have far more seats than any other institution. So if you can land a client like Bank of America, the number of people you actually reach is much larger than if you simultaneously signed 10 of the top individual portfolio managers, each with only 10 investors on their team.
Patrick O'Shaughnessy: If I think about institutions like Bank of America, everyone wonders how far along enterprise AI adoption actually is. Maybe your sample is a bit biased because you know your clients are using Rogo and using it a lot. But can you tell us where you think we are? This question seems genuinely hard to answer precisely.
Gabe Stengel: I'd say most companies are seeing huge gains in individual productivity and are trying to figure out how to convert that into measurable firm-wide productivity. I talk to individual bankers at any bank where we're deployed...
Patrick O'Shaughnessy: That's great.
Gabe Stengel: Oh, they'll say, I'm 100x more productive than I was, right? You'll meet a managing director who says, Gabe, I sent a client a five-page document that used to require three days of back-and-forth with an analyst, and I just did it myself in 10 minutes. These bankers haven't done any analytical work in 25 years, right? They haven't opened an Excel file in 20 years, but now they can do it themselves.
The question is, where do these productivity gains show up? Did you win more deals? Did you actually complete more transactions? Are you serving markets you didn't before? That's no longer just an individual productivity tool, it's a company strategy question. What's your plan? Are you using this to cut costs? Or to enter markets you previously thought weren't worth serving? You can look at banks like JPMorgan, which just announced it will try to do more M&A for small and medium businesses — markets they previously thought weren't worth serving because the deal fees might be too low and too many people were required. But now, if you have one banker who can handle the entire deal team's workload, you might be able to enter markets that never made sense before.
Patrick O'Shaughnessy: So in some sense, at some point the bottleneck becomes the client's own creativity, rather than your ability to supply unlimited capability. Soon you'll be dependent on them figuring out the answers themselves.
Gabe Stengel: The key is them figuring out what they want to do, right? If 100 great investors showed up tomorrow to work for you, how would you direct that productivity? You'd need time to figure out: what's the structure? Do they each handle different things? How much capital do I give each person? What directions do we attack?
Patrick O'Shaughnessy: Really interesting. Two years ago, the most obvious question on this topic would have been about accuracy, and people would use the word "hallucination," but that word seems to have disappeared from the conversation. I can't remember the last time someone said "hallucination" to me. Can you tell us, on the question of accuracy, in scenarios where precision matters down to the decimal point, what's the nature of this problem now?
Gabe Stengel: It's still extremely important. I think auditability is actually more important than accuracy. Of course, they're related. But what really matters is, I give you an answer and you know how to use it. If it's inaccurate and you don't know how to check it, and it's hard to see where it came from, then it's unusable whether accurate or not, because you don't trust it. If it's accurate most of the time, but even when it's occasionally wrong, you can easily see the assumptions and where the data was pulled from, then it's still operable and still saves you time.
And as these tools evolve from copilots you use just to retrieve information, into autonomous tools you trust not just to gather information but also to execute, to be autonomous, to actually make investment decisions or send emails, you need to have real confidence that if you go back and check why it made a decision, you can understand the reason, because you need to be able to debug it. Just like an individual investor makes bad decisions and you need to go in and see what went wrong — was it a bad data input? Was someone misleading it? What happened? You need to do the same with an agent. And especially in capital markets, where regulators are watching and need to ensure no violations occur, if you can't explain why a decision was made and how the data was input, that's not going to work.
Patrick O'Shaughnessy: You talked earlier about having to sell like a normal enterprise sales organization, but you're able to grow several times faster than the fastest-growing enterprise SaaS companies of the previous era. How do you solve that? You're somewhat constrained by human speed in enterprise sales. How do you hire enough people fast enough? How do you think about growing at the right velocity without API-usage growth like Anthropic or Cognition — their 10x growth is so easy, yours is much harder. How do you solve that?
Gabe Stengel: The core problem we need to solve is: how quickly can you make a person productive as a seller? Not just sellers, but marketers, sales development reps, post-sales people — how quickly can they understand our business, help move us forward, move clients forward, help our end users. So enablement, training, and continuous retraining is the fundamental problem we, and I suspect other fast-growing enterprise startups, have to face.
Chapter 7: Rogo's Corporate Brain
Patrick O'Shaughnessy: Do you use AI to build tools?
Gabe Stengel: Oh, our internal tools are amazing. First of all, every conversation that happens inside Rogo is recorded, 100% of the time. Whenever someone new joins, we say: "Hey Patrick, heads up, this is how we operate. You're recorded at all times and fed into the company brain." It's not a Big Brother surveillance thing. It's just that everyone you talk to will have Granola or some meeting transcription tool running because they need it, they need great information recall, they need to keep accumulating their knowledge. As a result, we've built up a massive knowledge base, and on top of that we've built various tools people can use, like: "We're working with this type of public equity company, facing this market — have we dealt with data like this before? Is there a case study that could be useful?" You can pull up a conversation from a colleague three weeks ago and you've never even met that person because they're based in Asia Pacific. Being able to absorb all of that information and retrieve it when people need it — that's the core of enablement.
Patrick O'Shaughnessy: Can you detail this internal brain?
Gabe Stengel: It's called "Shrek," for the simple reason that my engineers thought the name was hilarious. You can actually see it, there's a dashboard where you can check what people are working on — it's like a "swamp" — and it connects to all our different systems. It knows very clearly what our company's goals are — our values, what we want to deliver to clients, our north star metric. So it shapes every answer and every deliverable based on that. It can both push and pull. Someone can go in and ask: "Hey, I want a quick briefing on how I should introduce the topic of model routing to a large institution, articulate the value proposition, and the cost savings it can generate." It will pull up all relevant information for you. But it can also proactively say: "Hey Patrick, I see on your calendar you have a meeting Thursday with a private credit company. Here's everything you need to know, the use cases that would resonate most, and the ROI metrics that similar companies we've worked with in the past care most about."
Patrick O'Shaughnessy: What's your pitch for talent? For example, imagine someone who, if you could hire them tomorrow, would be disruptive because they're super capable or very famous. How would you convince them to come to Rogo instead of other equally exciting places?
Gabe Stengel: It always depends on that person's motivations, so it's hard to give a one-size-fits-all pitch. But the Rogo story I tell talent today is this: AI will completely transform the world. The most interesting place to be is applying AI, because that's where AI meets humans. So the companies that decide how AI intersects with humanity, how it reaches people, will be doing the most interesting, most creative engineering and product work in the world.
Finance is the catalyst for all human progress and innovation. Capital allocation sits upstream of funding every company, every idea, every economy. So if you can make that process more efficient, you can accelerate the entire world. And we are the leading player in this arena. Not just a $10 billion business, but a $500 billion business that completely transforms capital markets. There's so much depth here, such complex challenges, and so many interesting problems. It's incredibly exciting. And we have a group of exceptional people — smart, driven, curious, humble — who are going to get this done.
Patrick O'Shaughnessy: Nice pitch. Glad I joined, glad I invested. Talk more about capital markets. I do think that historically, the more efficient and liquid markets become, the more their positive impact grows. You can see this through market history, which isn't that long really — maybe three or four hundred years of true markets. What do you think is possible? Where could this go? Why do you believe — I'm guessing — that reducing friction in capital markets has this much power?
Gabe Stengel: When you say "finance is good for the world," there's always the risk of sounding like one of those billionaire PE guys saying "private equity is good for the world." But I like to look at the origins of high-end finance. When you think of a company like JPMorgan, one of its origins is J. Pierpont Morgan helping European investors connect with entrepreneurs in an emerging market — the United States — financing railroads and infrastructure that made America an economic powerhouse. This happened because intermediaries helped connect risk-taking capital allocators with entrepreneurial, innovative people. That had a profoundly positive impact on the world.
Now think about every corner of the economy — not just domestically in the US, but internationally — where capital markets are inaccessible. Every emerging country where it's very hard to raise financing to bring your ideas to life. Plus the 300,000 small and medium businesses in America that can't even articulate what Goldman Sachs does, and JPMorgan doesn't have time to work with them because the deals are simply too small. If you can accelerate the speed at which entrepreneurs, company founders, and individuals can access capital markets, you accelerate all of innovation.
Chapter 8: Chewing Glass
Patrick O'Shaughnessy: What have you learned from peers who are building companies similar in form to yours but in different industries?
Gabe Stengel: I've learned how much aggression it takes to grow this fast. I've learned what it feels like to chew glass every day, and how much belief you need in the long-term TAM and SAM to keep yourself from getting overwhelmed by all the problems that happen daily. You know, I call Winston of Harvey. Winston's ability to be unaffected by all the various "flesh wounds" that come at him at any moment, to focus only on the ultimate goal of where he'll be in three years and the two things that matter to get there — it's truly remarkable.
Patrick O'Shaughnessy: What does the "glass" look like? And what does "aggression" mean?
Gabe Stengel: I started working during COVID, so I never even got to see what an office was like. It was a very… you know, my co-founder John and I used to joke that it was an extremely expensive business school education. We got almost everything wrong. We didn't know how to hire, how to fire, how to manage people, how to give feedback, how to set direction. And the sheer number of people problems that arise during rapid scaling felt like chewing glass to me. So when employees leave, when you have retention problems, when you spend 6 months recruiting a candidate and they don't join — that's chewing glass. When you spend a lot of time on a product and it gets completely outclassed by the next model release, making you feel like an idiot for betting all your time and energy on a wrong decision — chewing glass. When you get rejected by 40 investors in a row before you finally raise a round — chewing glass. My experience of entrepreneurship is like a roller coaster. You have to feel the extreme highs and the extreme lows. I'm a very emotional person. I try not to let the team feel it, but at the peaks I feel like I'm on top of the world, and at the troughs everything feels catastrophic. But if I look back on this journey, the troughs get lower and the peaks get higher. When I look back at a trough or peak from three months ago, I think: what a joke, I could handle that with my eyes closed now. I think the key is regulating these emotions, channeling them into moving the business forward, but also not letting them distract you.
Patrick O'Shaughnessy: What about this aggressiveness thing? It feels like that's a common refrain.
Gabe Stengel: Pat Grady, whom you've interviewed, attended our board meeting on Wednesday. We presented a plan — whether hiring targets, business targets, or product targets — that was, to put it mildly, an extremely aggressive plan for next year. One of the things I love about Pat is his ability to distill everything into two bullet points with impeccable logic. He's like that: Premise one, premise two, conclusion follows. He said: "Gabe, if every person in the financial industry is going to make an AI purchasing decision in the next 18 months, and they're definitely going to buy something regardless — even if you're not in the room, they're going to buy — then the only thing that matters is that you can sweep the market as fast as humanly possible to make sure you're in the room." Given that, do you think this plan is aggressive enough? The answer is: No. It wasn't aggressive enough because I was being too soft. I think the reality is, you have to be aggressive enough to take all these risks, knowing clearly what game you're playing. As a venture-backed company, my goal is to widen both tails of the distribution. If the probability of failure goes up 30%, but the probability of becoming a $100 billion company goes up 20%, that's absolutely worth it. But you really have to accept that you're widening both tails at the same time.
What was your emotional lowest point when you were raising your Series A? At that time, we didn't have any marquee investors on the cap table, but we had one excellent investor — David Tish of Box Group, our early seed investor — who introduced us to a group of top investors to talk about the Series A. I thought, great, I've seen firsthand what having top investors on the cap table can do — for recruiting talent, for brand and customer attraction, and so on. This was finally our chance to achieve that. We'd been heads down building for two and a half years. David introduced us to 40 investors. I met all of them. Sequoia, Kleiner, Benchmark — all of them. All 40 rejected us.
And it wasn't the kind of rejection where someone gets an email and a deck and just isn't interested. It was the kind where — "Oh, this is interesting, let me meet Gabe. Oh, I like Gabe, let's talk with him for another hour. Oh Gabe, come to the investment committee. Oh Gabe, let's get dinner. Oh Gabe, come over this weekend. Oh, you know what? We've decided not to invest." It was very painful because at that stage, it all comes down to you personally. There's nothing else. It's like being dumped by 40 girlfriends, and each time you had real feelings, and each time they said: it's not you.
Thrive, similarly, spent a lot of time with me during the Series A. I had dinner with Avery and Vince. I fell in love with them and the firm because they were truly great. And then — the fatal blow — they also passed.
Fortunately, Keith Rabois appeared about a month after everyone had rejected us. Keith said: "Gabe, this direction isn't that contrarian really. It's basically Harvey for finance. Why should I invest?" I said: "Keith, if this isn't contrarian, why did all your friends just tell me it's a bad idea and not believe in me?"
Why do you think they didn't believe? There are many reasons. I think people underestimate the market size of finance, which is frankly ridiculous. But partly, I think San Francisco lacks intuition about the financial industry — they haven't invested in companies like Ion Group, Bloomberg, S&P, FactSet, Pitchbook, Morningstar. So there have actually been no truly great venture-backed companies in this market, and people don't have intuition for its shape and size.
The second reason was the product was too bad. Every investor said, "I work in finance, let me try it, will it change how I work?" They tried it and half the time the results were wrong, so they said "this can never work." My point was, we're on an exponential growth curve — in 6 months, 12 months, 18 months, it will work. I'll make it work.
The third reason was, they didn't believe I could pull it off. They didn't have enough data points watching me grit my teeth and keep grinding out the next iteration.
Patrick O'Shaughnessy: What do you think changed after that? Because now, now there's a very capable group of people in it.
Gabe Stengel: Because I got to know these people early, and I meet them every round. Every time, I say, hey, we're going to do this thing, and they say it's impossible. But every time, we do it. By the way, a lot of the time results manifest in different ways. Like, you actually lose key employees you needed, or that marquee customer you thought you'd land completely falls through. But we've kept finding ways, kept adapting to the market flexibly.
I think that's what you need to consider when investing. In a market like today's, especially in applied AI, where things are so turbulent and ambiguous, you need to bet on founders who can maintain extreme adaptability.
Patrick O'Shaughnessy: Why do you think there aren't more truly competitive financial services competitors?
Gabe Stengel: I think distribution is genuinely hard to break through. It's more of a people problem than an engineering and product problem — how to build trust, how to deliver value, how to work with these institutions. But at the same time, the engineering and product bar is extremely high. So the execution standard required to truly break into this market, I believe, is quite high. I've been very fortunate that some of the people I recruited early had finance backgrounds and were top talent. We recruited from our own networks people who had worked at Goldman, Citi, Jefferies, Apollo, Ares, or Blackstone. So we have an extremely scrappy, curious, smart team that is also made up of good people — low-key, humble, young enough to be open and hungry to improve.
Patrick O'Shaughnessy: I'm also very curious how you think about technical talent, and what matters most in a technical person on your team, and how that's evolved. There's an intuition that as execution becomes dramatically easier, the value of domain expertise in technical people rises. You no longer need to be as technically strong to write good code, at least not as much as before. So how has the shape of your engineering or technical team changed over time? I ask because I think about other companies that want to enter this space, regardless of their industry.
Gabe Stengel: Yes, I do think there are problems where the value of domain expertise is rising, where product intuition and a "general manager mindset" rather than a "pure engineer mindset" become extremely important. Of course, certain parts of our product require pure, resilient engineering because you're scaling very aggressively. But if I look at the people on the team who perform best, many are former founders — people who started companies, persisted, had strong product intuition, knew how to leverage it, and their companies may have eventually fizzled out.
As you said, there are many companies trying to crack financial AI. Many highly product-minded and engineering-minded domain experts have tried. We've acquired six different early-stage financial AI startups. Having former founders who can have a macro vision for the product's future, design what the user experience should be, while also having the engineering chops to keep making the right decisions — that's crucial.
Chapter 9: AI-Native Finance
Patrick O'Shaughnessy: You told me before we started recording that this is the first time we're seeing a true "innovator's dilemma" in this space. Can you explain what you mean?
Gabe Stengel: Yes. I think for many investment firms, for many high-end financial services institutions, the past 10, 20 years have been quite good. You know, being a capital allocator, an investor, you can make a lot of money. The barriers to entry in this industry are very high. Especially if you're in private markets, your business has a natural inertia — because you raise one fund, then another, and after that it's actually pretty hard to screw it up. I'm sure there are a hundred people who have raised funds who would say it's actually 100 times harder than I'm saying. But to be fair, the market hasn't had that shock moment that makes every investment firm, every bank slap their forehead and say, "Wow, I need to completely rethink what I'm doing." That moment of realizing — there are going to be hundreds of AI-native disruptors, AI-native investment firms, AI-native investment banks attacking my business model, and I need to figure out what I'm doing fast. Private equity firms, hedge funds, investment banks haven't faced a true "innovator's dilemma" in a very, very long time.
Patrick O'Shaughnessy: What does "AI-native" mean to you? What's the definition?
Gabe Stengel: In some sense, I think it means being willing to constantly re-examine and disrupt everything you do, and being so committed to AI that you stop worrying about what's possible or how crazy something sounds. You're just moving in one direction: integrating this extraordinary underlying technology into everything you do. It permeates every corner of the business. No business unit is sacred. No place is exempt from being completely transformed.
Patrick O'Shaughnessy: How do you actually do that concretely? You're a company leader who believes deeply in AI and wants to make sure — as you say at Rogo — that it's put into practice. But I'm sure there are areas where you're unsatisfied, where you feel AI isn't being used enough in some things. As a leader, how do you make sure the company treats this as an ongoing habit and practice rather than a one-time initiative?
Gabe Stengel: Let me give you something very specific: every month, I receive a report listing every person's usage of various AI tools across the company — tools we've purchased, tools we've built internally, all of them. Then within each department, I have a ranking of the top 5 and bottom 5 users. We post this ranking everywhere. The person who uses it least in each department gets a notice with a picture of a "dunce cap" posted around the office. It's funny, but everyone knows it's coming. And to be honest, as the company gets bigger, many people don't know me, don't know I'm actually joking, that I'm a funny guy, and they genuinely get scared. I hope they don't get truly scared, and I hope that dynamic changes over time. But regardless, it's a strong incentive mechanism that really ensures people use these tools.
Patrick O'Shaughnessy: If you list these problems — let's say someone is sitting there running this kind of company, doing well in the business, hard-earned but genuinely a good business. Honestly, these firms are often quite simple from an organizational and technical standpoint. It's mostly people, right? Not a lot of management overhead. Many of these firms do buy a lot of data and things like that. What questions would you encourage them to ask themselves and their companies to most effectively navigate this transformation?
Gabe Stengel: I'd say, if you're certain that today 90% of your company's enterprise value is in your people — your best investors, your best bankers, the ones who bring in the deals and revenue, where the value actually accumulates — and 10 years from now, the world's top investment firms and top banks will have 90% of their enterprise value not in people, but in software, data, and systems, then what do you start doing? You start figuring out how to take everything that's latent in the minds of your best people and convert it into a system that you own and that you operate. That's the first big thing.
Second, I'd recommend breaking down every step of the deal life cycle and, at each step, trying to clarify: where are you irreplaceable, or where do you have data that others don't, or domain expertise that others don't? Then honestly ask yourself: do I really have something in this market that others don't? Is it relationships? Is it background information no one else has? Or am I just assuming I'm smarter than others in this space, that I've read more? Because if it's the latter, AI is going to make that irrelevant.
Patrick O'Shaughnessy: Have you seen a major institution that's the most forward-thinking example of this attitude? Is there a person you love to cite as the example — the kind of — oh yeah, there are a few doing this.
Gabe Stengel: Interestingly, the people who look most visionary today sounded like madmen two years ago. The kind of person who comes in and says "we need to record every single conversation." You know, they say, can you feed this conversation directly into my company's brain, there will be a digital clone of me, and it will output game-theoretic analysis of how this investment should actually be made? Two years ago, when people heard people like this talk, they were thinking: what the hell is Patrick talking about?
One example is a co-founder, Molas John Momtazi, who is extremely prescient about where the industry is heading. He wants to create digital clones of all his firm's best bankers. He wants a system that can essentially appear on his behalf on phone calls, speak the way he speaks, understand his thinking, absorb the context. He wants to build a system where any junior employee in the investment bank can instantly access his expertise, his context, his relationships — so that information and data isn't just driving his personal revenue-generating ability, but enabling every junior banker in the firm to generate revenue.
So yes, there are people who have had their own "move 37 moment" and suddenly realized: wow, this is going to be far more profound than anyone expects.
Chapter 10: What Humans Are Still Better At
Patrick O'Shaughnessy: What are you most uncertain about regarding your business's future?
Gabe Stengel: I'd say it's the actual speed at which private markets transform. Right. If you think about why transaction volume and liquidity increase across different asset classes, it's often tied to standardization, because that makes assets easier to track and trade. Private markets have been immune to standardization because there's so much unstructured data. AI should solve that. That said, whether there will be regulatory or market forces pushing for additional standardization that truly accelerates transparency and liquidity in private markets is somewhat out of my control.
The other thing is, how much excess return remains in human relationships — that's still unclear to me. I'm talking about small-cap M&A. It's hard to imagine that if you're a small business owner who spent 20 years building a company, and you want to ensure that when you hand over the business, you trust the person, you can shake their hand — that process being fully automated. But for many sponsor-owned businesses, or things like secondary markets, private credit, GP/LP secondaries, I do think those can be fully automated. But those long-tail small businesses, mid-sized businesses, the ones dealing with generational transitions — how long it takes them to adapt to clicking a button to sell their business rather than shaking someone's hand — that's somewhat uncertain for me.
Patrick O'Shaughnessy: If there are a few young new founders here who are curious about raising from private market investors and interacting with them, what would you teach them? What would you tell them to do and not do?
Gabe Stengel: My fundraising style isn't for everyone. I'm very direct, very transparent about what worries me and where I want to go. But at the same time, you have to be extremely firm on that ultimate goal. Then I'd say it's about compounding and relationships. The people who suddenly appear and lead your Series C or D are often people I met at seed, then Series A, B, C. They passed each time for various reasons, but they accumulated more understanding of me and the business.
Is there anything you'd encourage people not to do? I think there's a lot of "fake it till you make it." You need to have that courage and confidence even when you're not fully confident. I'm an extremely anxious, extremely insecure, extremely fearful person. But you need to project confidence, to say you're confident about the direction you're heading, even when you know you're at a low point. I think people sometimes misunderstand this, thinking they need to pretend to be something they're not. I don't think that's right. You need to believe in that 5% chance — that you have the possibility of building a $100 billion enterprise — and that it's possible. You don't need to pretend it's 100% certain, but you should be able to clearly articulate a roadmap and strategy for how you could plausibly become a $100 billion company, and then be confident that if things go right, you'll get there.
Patrick O'Shaughnessy: Why are you fearful and insecure?
Gabe Stengel: I'm deeply anxious about everything that could go wrong. Every day feels like walking on a knife's edge with a thousand things that could collapse at any moment. I think this is the reality of building a business — it's a game of accumulating momentum. How do you do every small thing right so you can move downhill a little faster? I'm just afraid that momentum stops, or hits an obstacle, gets knocked off track, and then you have to regenerate it. I see clearly how hard it is to build a machine that keeps accumulating momentum. So if there's any stumbling block that could interrupt that momentum, if I'm not constantly vigilant about those moments, I won't be doing everything I can to prepare for them and make sure we can handle them.
Patrick O'Shaughnessy: What have we missed? What you've learned building this kind of company in this era — where everything feels like a scramble for possession, where it feels like every possible vertical will have a Rogo or Harvey-like competitor, especially where the "last mile" connections are hard to make. And where the dominant player won't just be Anthropic, won't just be OpenAI — it won't be one company ruling them all. So, what do you think is important about the experience of building this business right now that we haven't touched on?
Gabe Stengel: I think there will indeed be companies that solve all these problems, but the companies that can do this will become "black holes" that attract talent, capital, and brand — they'll be able to bring together resources to truly execute. Because AI is a strong tailwind, but to compete now, the bar for execution is higher than ever. Everyone has been pulled into the big leagues and is experiencing the "welcome to the NFL" moment. You have to be faster, stronger, more resilient, more agile than you anticipated. Having the right team matters more than ever. And the right team is extremely expensive these days. So if you can't figure out the strategy to become a black hole for talent and capital as quickly as possible, I think your risk of getting steamrolled by some lab or a company that's capable of doing this is much higher.
Patrick O'Shaughnessy: Every time I do these conversations, I save my favorite question for the end. What's the kindest thing anyone has ever done for you?
Gabe Stengel: I think I've benefited enormously from my parents. They're both incredibly kind, generous, and selfless. But their kindness manifests in very different ways. My mother's kindness is that no matter what I did, in her eyes I was remarkable, smart, could never do wrong — even though growing up, that was definitely not the truth. But she planted a seed of confidence in me that let me believe in myself, even in those low moments when everything felt like it was going sideways. Forty consecutive failures at something, failing again and again, failing tests, whatever it was — she acted like I might be the smartest person in the world. It was irrational, of course. But you need that irrational confidence that comes from unconditional love.
My father was completely different. If I came home having done something wrong, he'd be furious, barely want to look at me, completely unable to understand, because he's an extremely disciplined, principled person with extremely high standards. So his kindness was expressed in his willingness to try to understand who I was and why I might have failed at something, and then help me. He would sit down with me and go through every detail carefully, even though he sometimes genuinely couldn't understand why I had those opportunities and didn't seize them. He took the time to make me better, while also holding firm to his principles, his standards of excellence, and his definition of "good."
Patrick O'Shaughnessy: You're building a fascinating company and it's been really cool to watch it progress. Thanks for doing this with me.
Gabe Stengel: Thank you.
This article originally appeared from the WeChat public account "AI Native Lab", which continuously breaks down real-world AI implementation cases and shares enterprise AI practices and methodologies.