Gabe Stengel - Building Investing Superintelligence - [Invest Like the Best, EP.492]

1. Key Themes

The Model Eras Defined Product Capability, Not the Other Way Around

Rogo's usability tracked almost perfectly with foundation model releases rather than internal engineering breakthroughs. Stengel maps the company's entire history onto model generations: "I actually tried starting Rogo two times before we got started... nothing worked at all" 00:03:08, and then, "since things actually started to work, the eras are very tied to the model eras, right? It was 01 Pro and then probably Opus 4.5... with Opus 4.5 and the end of last year... the models just became capable of basically anything a junior investment professional or junior banker was doing, as long as you gave it the right instructions and context" 00:03:38. This implies vertical AI founders are making a bet on where models will land, and success is largely a function of correctly pre-building for a capability level that doesn't exist yet.

First-Mover Disadvantage in Applied AI

Building too early for a model capability that hasn't arrived yet actively hurts a company because early users form a negative impression that's hard to reverse. "There was a first movers disadvantage for a lot of applied AI companies because you thought you knew where the world was going and you wanted to build a product for it, but the models weren't quite there... you have a disadvantage" 00:04:04. The counter-insight: "if you were right about the end state and where the models were going and you were building towards that, when they get there, it's magical" 00:04:32.

Harness and Plumbing Matter More Than Model or Data

Stengel repeatedly argues the actual differentiator isn't proprietary data or a fine-tuned model but the surrounding infrastructure — compliance workflows, systems-of-record integration, and UX. "It's actually pretty interesting because we have to build a go-to-market machine five times faster than most enterprise sales organizations" and, more fundamentally, "the way that you harness these models is so, so important... intelligence... is not just because we have high raw recall and IQ and knowledge, but there's all these different microservices in your brain" 00:19:42. He points to Claude Code vs. ChatGPT as proof: "the models were actually fairly similar, but the harness and the way that it was presented from Claude was far better" 00:19:42.

Vertical AI Companies Must Be Perpendicular to the Labs, Not Parallel

The strategic bet is to build in the "messy," regulatory-heavy, high-TAM crevices that OpenAI/Anthropic will never bother with. "You want to build things that are perpendicular to what they want to build... there's a whole bunch of stuff underneath the surface that the labs are never going to build" 00:17:24. He frames the calculus explicitly: solving finance-specific workflow depth to reach "$5 billion in revenue" is "for Anthropic... kind of like stopping on the side of the road to pick up a penny" 00:17:50 given their trajectory toward a trillion in revenue.

Private Markets Are the Better AI Wedge Than Public Markets — Counter to Naive Intuition

Despite public equities seeming like the obvious AI target (clean, available data), Stengel argues private markets are structurally better because they are undigitized. "Part of the reason private markets are so attractive is because it's all done by humans, the coordination, the standardization, looking into things and the actual transacting. Whereas public equities, a lot of it has been automated" 00:11:49. The broader principle: "a lot of the opportunity to build an AI business in a vertical is somewhere where there's lots of plumbing that's not yet built" 00:11:37.

Pricing Models Must Evolve From Seats to Usage to Outcomes

Stengel lays out a three-stage pricing evolution for AI-native software. Rogo is currently per-seat because "our buyers are used to pricing per seat. They think of us in a similar category to Bloomberg, to FactSet" 00:25:03, but he wants to skip the usage-based/token-consumption stage entirely: "what if I just charge you for every good investment idea I give you? Or what if I charge you for the quarterly report you send to LPs that I can do perfectly?" 00:26:24. He frames token-based pricing as actually a disadvantage in some verticals: "There's a lot of industries where riding the coattails of token consumption isn't going to work for enterprise sales, and we're one of those" 00:25:28.

An Unprecedented Innovator's Dilemma Is Hitting High Finance

Historically insulated, momentum-driven finance businesses (funds, banks) haven't faced a real innovator's dilemma before — until now. "There hasn't been a moment and a shock to the market where every investment firm, every bank is saying, oh, wow, I need to completely rethink what I'm doing. And now there's going to be an opportunity for hundreds of AI native disruptors... to attack my business model" 00:50:55.

Individual Productivity Gains Are Not Yet Firm-Level Strategy

Enterprises are seeing massive individual output gains from AI but haven't figured out how to convert that into institutional strategy. "Speak to any individual banker... Life's great... 'I'm a hundred times more efficient than I used to be'... The problem is, where's that flowing through? Are you winning more deals?" 00:32:10. He cites JPMorgan's move into SMB M&A as the first visible example of a firm-level strategic response: "if you have a banker that can be a deal team of one, well, maybe you can enter parts of the market that before just made no sense" 00:33:04.

Auditability Trumps Raw Accuracy

The hallucination conversation has been superseded by a more nuanced framing around traceability. "It's actually more important to be auditable than it is to be accurate... If it's accurate most of the time, but even when it's not, it's very easy to see the assumptions that went in, where the data was pulled from... it's still actionable and it still saves you time" 00:33:52. This becomes existential as tools move from copilot to autopilot with real agency over transactions and regulatory exposure.

The Path to a Fully Automated Capital Markets Infrastructure

Stengel envisions AI collapsing capital-raising timelines by orders of magnitude, similar to the mortgage industry's digitization. "40 to 50% of mortgages are just delivered online by platforms like Rocket Mortgage... I think it'll take five minutes for KKR to figure out, can I sell this portfolio company to another sponsor? Not five months" 00:28:28. The endpoint is an "exchange" model modeled on Bloomberg's strategy but for agents: "I don't need to build the communication channel for humans to transact. I need to build the communication channel for the agents to transact" 00:30:08.

2. Contrarian Perspectives

"AI Native" Doesn't Mean Building AI Products — It Means Nothing Is Sacred

Most firms think of AI adoption as tool procurement. Stengel's definition is far more radical: "It just means willing to constantly reinvent everything you're doing and being so AI pilled that you don't worry about what's possible or what might seem completely farfetched... there's no part of the business that you hold sacred that is immune to being revolutionary" 00:51:19. His proof point is a competitor's co-founder who two years ago sounded "batshit crazy" for wanting digital clones of every banker recording every call — and now looks prescient 00:54:31.

The Founder Was Rejected by Nearly Every Top-Tier VC — And That Rejection Was Rational, Not a Signal of a Bad Idea

Conventional wisdom says 40 rejections signals a weak business. Stengel argues the rejections were reasonable given the information available at the time, not evidence the idea was flawed: "the product was terrible. And so every investor said, oh, I work in finance. Let me use it. Is it going to transform what I do? And they tried it and it was wrong half the time... 'this is never going to work'" 00:46:37. Keith Rabois only invested a month later once the trajectory was more visible, and even called it "not a contrarian bet... basically just Harvey for finance" 00:46:12 — underscoring that great outcomes sometimes come from being early on a correct-but-currently-unconvincing thesis, not from being contrarian in the classic sense.

Enterprise AI Sales Requires MORE Human Headcount Growth, Not Less — Contrary to the "AI Kills Sales Headcount" Narrative

While the popular narrative is that AI collapses go-to-market costs, Stengel argues his category requires building "a go-to-market machine five times faster than most enterprise sales organizations ever have to build" 00:25:28 precisely because buyers demand human-mediated, seat-based trust relationships — the opposite of the token-consumption self-serve growth of Cursor or Cognition.

The Biggest Risk to Incumbent Finance Firms Isn't Technology, It's That They Don't Know What's Actually Defensible

Stengel suggests most senior people at funds/banks conflate "I am smart and well-read" with genuine moat, when in fact only relationships and proprietary data endure. He advises firms to interrogate: "do I actually have something that no one else has in this market? Is it a relationship? Is it context that no one else has? Or do I just think I'm smarter and better read on the subject area in the market? In which case AI is going to obviate that" 00:53:53 — a direct challenge to how highly-paid professionals view their own value.

Fake Confidence Is Bad Advice — Founders Should Instead Believe a Small Probability Is Genuinely Achievable

Against the "fake it till you make it" startup trope, Stengel argues the correct posture is quantifiable conviction, not performative bravado: "You need to believe that that 5% likelihood that you can be a $100 billion business is likely. And you don't need to pretend it's 100% likely, but you should be able to delineate with a very clear roadmap and strategy how it is possible" 00:58:14.

3. Companies Identified

Rogo — AI platform for capital markets/dealmakers (banks, PE, investors). Stengel's own company; discussed at length as the case study for building applied AI in finance, from data-room automation to email markup workflows. "I'm saving hundreds of hours a month. I am doing things I never could have done before" 00:05:00 (describing customer feedback).

Anthropic (Claude / Claude Code / Cowork) — Foundation model lab. Cited as the benchmark for harness quality and as the "coattail" that Rogo's own spend on tokens has ridden. "The amount that we pay them has risen exponentially without a human loop because it's a token consumption model" 00:25:28. Also cited for harness superiority: "the models were actually fairly similar, but the harness and the way that it was presented from Claude was far better" 00:19:42.

OpenAI (ChatGPT) — Mentioned as comparison point to Anthropic's harness advantage, and as a lab that will never build finance-specific infrastructure like data rooms: "I don't think OpenAI or Anthropic will ever want to build a data room business" 00:18:56.

Jane Street — Cited as the historical analogy for how long it takes to build a dominant franchise around a technological edge: "Jane Street took 15 years to build the dominant franchise" in market making and quant trading 00:01:41.

Bloomberg — Referenced repeatedly as the strategic template Rogo is following: "get a little bit of data in the door, build all the analytics and workflows on top... and then provide the exchange and communication platform" 00:29:41, including Bloomberg Messenger as the analog for private-market agent-to-agent communication infrastructure.

FactSet, Capital IQ, PitchBook, Morningstar, S&P, Ion Group — Named as the incumbent financial data/software category that Rogo is positioned against/adjacent to, and noted as a reason Silicon Valley lacks intuition for the finance software market: "SF has less intuition for finance because they didn't fund Ion Group or Bloomberg or S&P or FactSet or PitchBook, Morningstar" 00:46:37.

Harvey — Legal AI company, run by Winston Weinberg, cited as a peer/model for founder resilience and long-term focus. "Winston's ability to just not worry about the hundred flesh wounds that are inflicted on him at any given time. And just think about the kind of end state goal of where he's going to be three years from now" 00:41:59.

Affirm — Cited via Max Levchin's practice of rebuilding core ledger technology annually as a model for constant reinvention. "To firm, they rebuild the fundamental ledger technology every year... it's the most interesting engineering problem... it's a good way to make sure that system doesn't ossify" 00:22:53.

JPMorgan — Cited as an example of a bank making a firm-level strategic pivot enabled by AI productivity gains: "JP Morgan just announced that they're going to try and do a lot more M&A work for SMBs, for parts of the market that before they didn't think it made sense to go out and serve" 00:33:04.

Rocket Mortgage — Used as the historical analogy for how human-intermediated, "important" financial decisions (mortgages) became automated over 15-20 years, foreshadowing capital markets: "now, 40 to 50% of mortgages are just delivered online by platforms like Rocket Mortgage" 00:28:28.

Robinhood — Referenced as the consumer-equities analog for what frictionless capital raising for businesses could look like: "someone who works at a company to go online and click a button and try to raise capital the way that someone can go onto Robinhood and click a button and buy an equity" 00:28:28.

Bank of America — Named as a large-scale distribution customer example illustrating why Rogo prioritized big banks: "if you can land a bank like Bank of America, you can actually get in the hands of far, far, far more people than if you land the 10 best single portfolio manager" 00:31:40.

Box Group — David Tisch's fund, credited as an early believer and connector who introduced Stengel to Series A investors despite the eventual wave of rejections.

Thrive Capital — Cited as a firm that spent significant time with Stengel and ultimately became a believer/investor relationship: "We went to dinner with Avery and Vince, fell in love with them and the firm because they were awesome" 00:45:44.

4. People Identified

Gabe Stengel — Founder/CEO of Rogo; former investment banker (buy-side M&A coverage) for a "handful of years." Central figure of the episode; built Rogo through two prior failed attempts at similar ideas before finding product-market fit aligned with model capability jumps.

Dario Amodei (referenced as "Dario") — Anthropic CEO. Cited for his framing of AI's trajectory: "everyone's going to have a data center full of geniuses or a country full of geniuses in the data center" 00:02:05.

Winston Weinberg — Founder/CEO of Harvey. Praised for founder resilience and focus discipline: "how much aggression it takes to grow this quickly... Winston's ability to just not worry about the hundred flesh wounds" 00:41:33.

Max Levchin — Affirm founder. Cited as inspiration for the practice of annually rebuilding core technology to prevent ossification and retain engineering talent 00:22:25.

Pat Grady — Sequoia investor and Rogo board member. Praised for distilling strategy into simple, "logically infallible" frameworks: "he boils everything down into like two bullet points... premise one, premise two. This is the result" 00:43:51. Pushed Rogo's board plan to be more aggressive: "the only thing that matters is that you can blitz the market as quickly as possible to make sure that you are there" 00:44:17.

David Tisch — Box Group investor; early believer who connected Stengel to 40 Series A investors despite the ultimately painful rejection cycle. Described as a critical early supporter 00:45:17.

Keith Rabois — Investor who backed Rogo roughly a month after ~40 other investors passed. Notable for correctly assessing the business as not even contrarian: "Gabe, this isn't a contrarian bet. It's basically just Harvey for finance. Why would I do it?" 00:46:12 — yet he invested anyway when others hadn't.

John Montazzi — Co-founder of Firm Mollus (financial services firm), described as one of the earliest people to see the AI transformation coming in finance and push for radical solutions (digital clones of bankers, embedded institutional knowledge systems). "He wanted digital clones of all their best bankers... he wanted to build a system such that anyone in junior levels of that investment bank could leverage his expertise" 00:55:05.

John — Stengel's Rogo co-founder, referenced regarding early operational mistakes during COVID-era company building: "John, my co-founder and I used to joke that it's a very expensive business school education" 00:41:59.

Lee Sedol (referenced as "Lisa Dahl" — transcript garble for the Go world champion, related to AlphaGo's "Move 37") — Used as the reference point for a potential paradigm-shifting moment in investing analogous to Move 37 in Go, where AI could exhibit judgment beyond human comprehension 00:12:33.

5. Operating Insights

Build an Internal Company Brain That Records Everything

Rogo records every internal conversation into a searchable, prescriptive knowledge system ("Shrek") that encodes company goals, values, and metrics, then proactively pushes relevant context to employees before meetings. "It's connected to all of our different systems. It is very prescriptive about what it knows our company goals are... it can also say, hey, Patrick, I see on your calendar on Thursday, you're meeting with this sort of private credit firm. Here's all the information you should know" 00:37:28. This directly solves the enablement/onboarding bottleneck that Stengel identifies as the core constraint on hypergrowth sales orgs.

Publicly Stack-Rank AI Tool Usage With Social Pressure Incentives

To force AI adoption as a habit rather than a one-time initiative, Stengel built a monthly leaderboard of AI tool usage by division, with public "dunce cap" printouts for the bottom users. "I have a stack ranking within every division of the top five power users and the bottom five users. And we post that everywhere. And the bottom user in each division gets a printout with a dunce cap and we post it around the office" 00:52:15.

Acquire Fledgling Founders as a Talent Acquisition Strategy

Rather than only hiring engineers, Rogo systematically acquired six different failed/struggling financial AI startups specifically to bring in ex-founders who combine product intuition with engineering chops. "There's been a lot of really smart, product-minded, engineering-minded domain experts who have tried to tackle finance AI, and we've acquired six different fledgling financial AI startups" 00:49:47.

Constantly Rebuild the Core System Before You Hit a Local Minimum

Rather than waiting for stagnation to force a rebuild, Stengel applies Affirm's ledger-rebuild philosophy proactively to Rogo's agentic harness: "We are constantly looking at it, realizing we're not even at a local minima. It would be impossible if we were at a local minima because the models are changing so quickly and we need to redo the whole thing" 00:22:53.

Evaluate Vertical AI Opportunities on Three Specific Founder/Market Filters

Stengel's explicit investment framework for vertical AI: (1) enough industry complexity/data depth that a simple ChatGPT wrapper can't win, (2) founding team domain expertise and obsession, and (3) willingness to "constantly slash the product to nothing." 00:21:00 This is a reusable diligence checklist for any vertical AI thesis, not just finance.

6. Overlooked Insights

The Regulatory Discoverability Time Bomb

Buried in a discussion about compliance, Stengel drops a detail with enormous downstream implications: "if someday some Delaware court judge makes AI inputs and research discoverable, you have actually done it all the right way such that you're not intermingling information because the reality of AI investment judgments and AI banker outputs is that you're going to be able to see the full lineage of how those things are created" 00:16:32. This implies an entirely new legal exposure category is coming for every fund and bank using AI in decision-making — full auditability of AI reasoning could become discoverable evidence in litigation, fundamentally changing how firms need to architect and retain AI outputs today, years before any court ruling forces the issue. Few firms appear to be building for this proactively; Rogo is quietly using it as a structural product requirement and competitive moat.

The Compaction Problem Will Explode as Agents Move From 1:1 to Many:Many

Stengel names "compaction" (how an agent retains and compresses relevant memory across interactions) as the single hardest unsolved technical problem — but frames it almost as an aside. The real insight is buried in his scaling point: "almost all agents are one-to-one... As soon as these are actually things that consort with a lot of colleagues or in a Slack channel with a hundred people... the compaction problem just scaled exponentially because it's having conversations with a hundred different people and needs to be able to coordinate across those things" 00:23:33. This suggests that the entire current generation of enterprise AI tools (built for single-user context windows) may need fundamental re-architecture once multi-agent, multi-human collaborative workflows become the norm — an infrastructure gap almost nobody is discussing publicly relative to model capability, yet it may be the actual bottleneck determining which applied-AI companies survive the next platform shift.