Why OpenAI and Anthropic Won't Win Finance
In a Nutshell
Rogo's founder claims OpenAI and Anthropic won't dominate finance because vertical AI companies can build the specialized "plumbing"—compliance systems, data room infrastructure, and workflow integrations—that labs won't pursue. The core bet is that in 10 years, 90% of investment firm value will shift from people to owned AI systems, with private markets offering the biggest opportunity since they're still human-driven while public markets are already automated. Success requires constant product reinvention, deep domain expertise, and building systems that capture what currently lives only in top investors' minds.
These notes were generated by AI and may contain inaccuracies.
Current AI models are smarter than anyone the speaker knows. The vision is that every portfolio manager at a hedge fund could have 10,000 agents that fraternize, talk about ideas, read through notes, and pontificate, then deliver one idea after 24 hours of endless debate. In 10 years, the world's best investment firms and banks will have 90% of their enterprise value in software, data, and systems rather than in people. The key question becomes how to take everything that lives in the latent minds of the best people and put it into a system that the firm owns and operates.
The goal is building investing super intelligence tools to help investors do their job much faster, better, cheaper, easier, and with higher quality. The trajectory has advanced to the point where core functions that even very smart analysts or portfolio managers were doing a couple years ago are now being automated. The two-year timeline is easier to reason about than 10 or 20 years because the best investors are going to spend the next 2 to 5 years figuring out how to integrate AI into what they do and reinvent their firms and themselves.
Jane Street took 15 years to build the dominant franchise in market making and quant trading. The question becomes what would Goldman, Millennium, or Citadel do if they had a country full of geniuses show up in a data center. The biggest challenge over the next 5 years for every great investor is figuring out how to apply AI into the investment life cycle.
The founder tried starting Rogo two times before successfully launching. In high school, attempted to build an app using old AI techniques to track equity exchange rates of two public companies during mergers, but it was terrible. In college before GPT-3 came out, published a paper on AI assistance for econometrics and financial econometrics, but commercializing at the time didn't work.
The actual business started when GPT-3 came out, pre-ChatGPT. The early days showed how magical it was in demos, but nothing worked at all. The eras of product capability are tied to model releases: 01 Pro was the first time there was enough reliability to function as a good search tool, reliably calculating financial metrics for businesses over the last 12 quarters without being so annoying that users would do it themselves.
With Opus 4.5 at the end of last year into the beginning of this year, the models became capable of basically anything a junior investment professional or junior banker was doing, as long as given the right instructions and context.
A great product is defined by feedback received daily from users saying it is transforming the way they work, saving hundreds of hours a month, enabling them to do things they never could have done before, making them smarter, and helping them make better decisions. The product brings joy in day-to-day use because the UX, attention to detail, and craftsmanship are built specifically for financial professionals.
The product makes it easy for a managing director at a bank to email a markup of a deck - which is how they typically do workflows anyway - except sending it to an AI analyst over email instead of a human analyst, then returning that markup in 20 minutes instead of 2 days. The system simultaneously alerts the junior analyst on the deal about what's happening and shows them the full auditability of all markups made. This UX flow makes adoption easier for financial professionals who haven't logged into a computer in 10 years but know how to mark up documents on an iPad.
The coolest innovation being worked on is multi-agent book scenarios. The reason this is possible is because investors are happy to pay $50,000 for one really good idea, whereas few other domains allow expanding that many tokens for one simple insight. The models are now smarter than anyone the speaker knows or spends time with. The cutting edge is the plumbing: connecting the system to context, informing it about investment theses, telling it the way users work, and integrating it into existing workflows. Building out all the plumbing to collect data and context is what is cutting edge.
Early users are dealmakers or people transacting - buying companies, selling companies, coordinating transactions. The system helps prepare data rooms, unpack data rooms, coordinate calls with third parties to discuss data rooms, and handle initial steps through closing of deals. Users are either on the sell side or buying side of transactions, using the system to prepare thoughts and materials to execute the full deal.
This includes putting things into a data room such as company models, PowerPoints describing customers, and answers to DDQ questions on customer concentration. On the buy side, agents tear through data and map it to the firm's investment philosophy to assess whether concentration is lower than the risk profile wanted for the fund.
People access the system via classic channels: email, chat bots, proactive alerts. The system also operates behind the scenes in firms' systems because when working on a deal, users need to update CRM systems, portfolio and monitoring systems, and distribute information to LPs. Half the surface area is the underneath of the iceberg - interacting with different systems of record based on what humans are doing over the course of deals.
The founder has not spent nearly as much time as should have with great public markets investors and does not know what it takes to be a great public markets investor. The company needs to hire domain expertise and figure out how to apply built systems to that market. There is extreme conviction that the underlying systems, tools, and infrastructure will be invaluable to public markets, but the company needs the right people to piece it together for the end state product and last mile delivery for public equities investors.
The board pushes to expand the ICP beyond core banking, but there has been so much depth in TAM in the dealmakers vertical. The end state vision is being the full infrastructure for private markets where people can transact very effectively, which is far more important to dealmakers. For public equities, all infrastructure and exchanges already exist. The founder wants to serve the most sophisticated, smartest users who have inordinate amounts of knowledge on companies and industries they track.
Private markets are attractive because everything is done by humans - coordination, standardization, looking into things, and actual transacting. Public equities have been largely automated. The opportunity to build an AI business in a vertical is where there is lots of plumbing that is not yet built, then applying tools on top of that.
The world's best investors have ways of figuring out what matters and exercising their own judgment across a range of topics. The jury is still out on whether AI can eventually replace that. When AlphaGo's move 37 happened, Lee Sedol saw something he could never see. If that starts happening in public equities, it will really change what matters.
The core skill set will be folks who can go out and gather data and inputs into their models that no one else will have. If investors can spend time in the field, speak to experts, and develop relationship graphs of people who can inform their models, they can feed their models with data that no one else has.
Early Rogo was a Rube Goldberg contraption with 60 different model calls. A question comes in, the system tries to identify what companies are being asked about, finds their tickers, feeds those tickers into API calls to Bloomberg or FactSet or internal data sets, then calls different models to pull everything together. As models get smarter, the approach shifts from being prescriptive to thinking about the simplest, best, highest quality tools.
The company spends time thinking through all data inputs that a great banker or investor would need, all compliance and regulatory requirements to ensure full lineage of how AI investment judgments and banker outputs are created, and benchmarking models to create evals and data sets to decide what is most performant, cheapest from a token perspective, and lowest latency, then routing tasks to appropriate models.
The strategy is to build things perpendicular to what OpenAI and Anthropic want to build. Sometimes building a chatbot helps go to market faster, but there is a whole bunch of stuff underneath the surface that the labs are never going to build that needs to be built for finance. Financial services and capital markets have many businesses generating more than 5 or 10 billion by going deep into workflows, data sets, and how work is done. Finance is a collection of different niches with different data sets, different definitions of good, and different regulatory requirements. The goal is reaching $5 billion in revenue by going deep across those areas and creating systems of record that help manage them.
Anthropic would view stopping to serve finance niches as stopping on the side of the road to pick up a penny because they are on the pathway from $100 billion in revenue to a trillion in revenue.
One example is ingesting MNPI when working on transactions or deals that affect the market. Compliance requirements include auditability, what can be flagged, and how information feeds into internal systems of investment firms or banks so that if audited or regulators want to see what was done, all plumbing is pixel perfect.
Another example is when a big public company buys another big public company and needs to send data back and forth, requiring a data room that is compliant, safe, and secure, coordinates transactions, and is plugged into the way work is done with agents and workflows rather than a static Dropbox folder.
The fundamental difference between Claude Code when it came out versus OpenAI and ChatGPT was that the models were fairly similar, but the harness and way it was presented from Claude was far better. This allowed models to exercise more of their long-running capabilities, leading to runup in usage and huge expansion. The way models are harnessed is so important. Intelligence alone is not why humans are high agency and can do a lot - there are microservices in the brain for putting knowledge away, retrieving it, and triggering things. Emotions trigger microservices, and great investors have good spidey sense for when they see something in the market that triggers recall from past events that informs creative decisions.
The industry must have enough complexity and depth in types of data, types of systems of record that people use, and types of deployment models that significant effort can be spent solving those problems to have a wedge to solve everything else.
For AI companies to build sustainable competitive advantages, the target industry must meet specific criteria. If an industry allows anyone to immediately sell basic chatbots or co-pilots without solving complex integrations, there isn't enough time to build the perpendicular capabilities needed for differentiation. The industry needs to be adjacent enough to the core market to allow meaningful competitive moats.
Founders don't need unique domain expertise to get started, but sufficient knowledge combined with genuine curiosity about the sector is essential. Growing up in New York surrounded by high finance culture provided foundational exposure, and assembling a team with deep domain experience becomes critical for execution.
The company has assembled over 100 people who have spent time within investment banks or investment firms across the world's best institutions. This enables the team to constantly take newly released models and harness them for finance applications, with the core job being to catch changes in the models and figure out how to apply them into institutions.
A critical success factor is building a business willing to constantly reinvent the core product and slash it to nothing when necessary. Delivery methods that cannot be fully cannibalized quickly, such as terminals or very specific UX interfaces, won't be agile enough to reinvent every six months when there's a step change in capabilities.
The approach draws inspiration from Max Lebchin's approach at his firm, where they rebuild fundamental ledger technology every year. This serves two purposes: it creates the most interesting engineering problems to retain talent and teach engineers about the core business, and it prevents system ossification while ensuring constant improvement.
The company applies this same philosophy to their harness and core agentic system, constantly recognizing they're not at a local minimum because models change so quickly that the whole system needs to be redone.
Compaction represents a significant current limitation. When users have 100 conversations with a single agent, the system must remember the right things and compact memory into usable tokens while maintaining coherence and context about the user. This compounds exponentially when agents work across many colleagues or in Slack channels with 100 people, as the system must coordinate across multiple conversations and maintain context over numerous interactions.
Two distinct categories of AI software businesses are emerging. The first follows typical enterprise sales models but uses AI models as a tailwind to build significantly better products. The second category consists of token brokers and token consumption businesses, where usage-based tools like Cursor, Factory, and Cloud Code can scale commercially with fewer people because customers are accustomed to buying usage-based tools.
The company operates as a classic enterprise software business pricing per seat, as buyers are accustomed to this model. They position themselves in the same category as Bloomberg, FactSet, Capital IQ, and PitchBook. Enterprise sales require significant human involvement - every deal needs an AE, solutions architect, and sales engineer for integration explanations. Unlike token consumption models where usage can rise 100-fold without human intervention, enterprise deals cannot scale that way.
Every business needs to progress through two pricing revolutions: moving to usage-based pricing, then to outcome-based pricing. The goal is to skip token-based and usage-based models entirely, moving directly to charging for specific outcomes like good investment ideas, quarterly reports delivered to LPs, or SIMs created for bankers.
Being only 1% into the product roadmap means 99% of capital markets innovation remains ahead. The priority is being a good partner and steward of customers' AI strategy so they want to continue working together in the future, rather than focusing on immediate pricing optimization.
The percentage of capital markets workflows, investment workflows, and investment banking jobs that remain completely human rate-limited is substantial. Similar to how mortgages transitioned from requiring branch visits to 40-50% being delivered online through platforms like Rocket Mortgage, capital markets will see dramatic efficiency gains.
Future capabilities will include: business owners raising capital in five minutes similar to buying equity on Robinhood, KKR determining portfolio company sales to other sponsors in five minutes rather than five months, and pricing assets in an order of magnitude less time. This will make markets more transparent, liquid, and efficient with significantly more activity.
The approach mirrors Bloomberg's strategy: offer minimal data to gain entry, build comprehensive analytics and workflows, then provide exchange and communication platforms. However, instead of building communication channels for humans, the focus is on building channels for agents to transact across businesses and investment firms.
The infrastructure needed includes systems allowing large private equity firms to have agents negotiate deals, correspond with third-party consultants, legal advisors, and run auction processes with multiple sponsors providing bids.
The customer base consists mostly of large banks due to the founder's investment banking background doing buy-side M&A coverage. Investment banks serve as the distribution channel for the rest of finance, as many great investors began as analysts at firms like Goldman Sachs. Banks also have the most seats, enabling broader reach than targeting individual portfolio managers.
Individual productivity gains are massive, with bankers reporting being 100 times more efficient. Examples include MDs creating five-page client deliverables in 10 minutes that previously required three days of back-and-forth with analysts. Bankers who haven't opened Excel files in 20 years can now perform analysis independently.
The challenge is converting individual productivity into measurable firm productivity. Banks like JP Morgan are exploring using AI to enter previously uneconomical market segments like SMB M&A work, where deal fees were too small to justify traditional staffing but become viable with single-person deal teams.
Auditability has become more important than raw accuracy. When answers include visible assumptions and data sources, they remain actionable even when occasionally inaccurate. As systems transition from co-pilot tools to autopilots with agency to execute investment decisions or send emails, full confidence in decision traceability becomes essential for debugging and regulatory compliance.
The core constraint is how quickly humans can become productive in sales, marketing, SDR, and post-sales roles. Enablement and continuous retraining represent the fundamental challenge for fast-growing enterprise startups.
All internal conversations are recorded 100%, with new employees informed that everything is captured and filtered into the company brain. This creates a massive information reservoir accessible through tools that can surface relevant past conversations and use cases.
The internal brain system, called Shrek, connects to all company systems and maintains prescriptive knowledge of company goals, values, and north star metrics. It functions both proactively, surfacing relevant information for upcoming meetings, and reactively, responding to specific queries about topics like model routing value propositions.
The pitch to potential talent emphasizes that applied AI represents where AI intersects with humanity, creating the most interesting creative engineering and product work. Finance serves as the catalyst for all human progress through capital allocation, which sits upstream of financing every company, idea, and economy. The opportunity exists to build not just a $100 billion business but a $500 billion business that completely transforms capital markets.
Historical analysis of 300-400 years of proper markets shows that as markets become more efficient and liquid, their positive impact grows substantially. The origins of high finance, exemplified by JP Morgan connecting European investors with American entrepreneurs to finance railroad infrastructure, demonstrate how intermediaries linking risk-takers and capital allocators with entrepreneurs create profoundly positive economic effects.
Current inefficiencies prevent 300,000 American businesses and emerging nations from accessing capital markets. Accelerating entrepreneurs' and founders' ability to tap into capital markets can accelerate all innovation across the economy.
Vanta automates compliance so teams spend less time on security reviews and more time getting customers. Vanta cuts audit prep by 82% and provides instant up-to-date proof of trust without manual work. Customers report a 526% return on investment and more than 16,000 companies use Vanta, including Ramp, Harvey, and Snowflake.
Ridgeline is the first end-to-end system of record with embedded AI for investment management firms, running portfolio, accounting, reconciliation, reporting, trading, and compliance on one unified platform. Firms are moving off legacy technology onto Ridgeline because of how far ahead Ridgeline's AI features are compared to anything else in investment management software.
Investment managers fall into two camps regarding AI: some unsure where to start and others convinced they can build their own order management system over a weekend. The reality is that running an investment firm will always require governance, controls, and a single source of truth for data. No amount of AI enthusiasm changes that requirement. Firms that come out ahead in the AI era will be running on Ridgeline's unified platform.
Gabe learned how much aggression it takes to grow this quickly and how much conviction is needed in the long term. He references Winston at Harvey, noting Winston's ability to not worry about the hundred flesh wounds inflicted at any given time and focus on the end-state goal three years from now.
Gabe worked during COVID and didn't see what an office looked like. He and co-founder John joke that it's an expensive business school education because they were doing everything wrong. They didn't know how to hire, fire, mentor, manage, give feedback, or set direction. People problems that arise with scaling quickly feel like eating glass. Examples include people quitting, retention issues, candidates not joining after six months of recruiting, products getting washed over by newer models, and getting rejected by 40 investors in a row.
Gabe describes startup building as a roller coaster where you have to feel extreme highs and extreme lows. He's a super emotional guy and tries not to let the team feel it, but he feels on top of the world at highs and like everything is cataclysmic at lows. Looking back, the lows get lower and the highs get much higher. Three months ago, problems that seemed significant now feel manageable.
Pat presented an extremely aggressive plan for next year in terms of hiring goals, commercial goals, and product goals. Pat boils everything down into two bullet points that are logically infallible. When asked if the plan was aggressive enough given that everyone in finance will make AI buying decisions in the next 18 months, Pat said no because Gabe was being soft. The goal as a venture-backed business is to increase the tails of the distribution - it's acceptable if there's 30% more likelihood of failure if the odds of becoming a hundred billion dollar company increase by 20%.
When raising Series A, there were no star investors in the cap table yet. David Tish at Box Group introduced Gabe to 40 investors including Sequoia, Kleiner, and Benchmark. All 40 passed. The process was personal: investors would meet Gabe, like him, spend an hour with him, invite him to IC, have dinner, then pass. This felt like being broken up with by 40 girlfriends. Thrive spent significant time with Gabe, went to dinner with Avery and Vince, then delivered a crushing blow. Keith Rabois came a month after everyone else rejected them.
Keith initially said this wasn't a contrarian bet but basically Harvey for finance. Gabe countered that if it wasn't contrarian, why did every single investor say it was a bad idea? The reasons investors passed included: people underappreciated the TAM in finance, partly because SF has less intuition for finance since they didn't fund firms like Ion Group, Bloomberg, S&P, FactSet, PitchBook, or Morningstar. The product was terrible - investors tried it and it was wrong half the time. People didn't think Gabe could figure it out, lacking data points of watching him chew glass.
Gabe met all these investors early and at every round. Every time he said they were going to do something, they did it. Sometimes they would lose key employees or excellent customers, but they kept figuring out what to do and navigating the market. When investing in applied AI markets that are turbulent and ambiguous, investors need to underwrite founders being extremely dynamic.
Distribution is hard to crack and it's more a people problem of building trust and delivering value with institutions than just an engineering and product problem. There's also an enormously high engineering and product burden. The standard to execute to crack this market is high. Gabe was lucky that early hires were ex-finance and killers, hiring from inner network including Goldman Sachs, Jefferies, Apollo, Ares, and Blackstone. The team consists of diehard, ambitious, curious, smart people who are good humans, low ego, humble, young enough to be open-minded.
Domain expertise is increasing in value, and product intuition with a GM-like mindset versus engineer-like mindset is important. Some product surface areas need raw gritty engineering talent because of aggressive scaling. Excellent team members are often former founders who started businesses, persevered, ate glass, had product intuition, and figured out how to channel it. Harvey has acquired six fledgling financial AI startups. Former founders who can be visionary about product future, navigate UX decisions, and have engineering chops to make constant right decisions are super important.
For investment firms and high finance financial services, the last 10-20 years have been good. It's a hard industry to enter, and private markets businesses have natural momentum from raising successive funds. There hasn't been a moment where every investment firm and bank needs to completely rethink what they're doing, creating opportunity for AI native disruptors to attack business models.
AI native means being willing to constantly reinvent everything and being so AI-pilled that you don't worry about what's possible or what might seem far-fetched. You chart a trajectory toward integrating this alien fundamental technology into everything, showing up in every part of the business with no part held sacred as immune to revolution.
Every month, Gabe receives a report showing how everyone at the company is using various AI tools - both procured and built internally. There's stack ranking within every division of the top five power users and bottom five users. The bottom user in each division gets a printout with a dunce cap posted around the office. It's hilarious but creates strong incentive function to use AI tools.
If 90% of enterprise value is currently in people (best investors, bankers bringing deals and revenue) and in 10 years the world's best investment firms will have 90% of enterprise value in software, data, and systems, what would you start doing? Take what lives in the latent minds of best people and put it into a system owned and operated autonomously. Unpack every part of the deal lifecycle and chart where you have invaluable data or domain expertise that no one else has, then be diligent about whether you actually have something unique or just think you're smarter.
The folks who look most prescient today sounded crazy two years ago. John Mumtaz, co-founder of a firm, was prescient and wanted digital clones of all their best bankers - systems that could show up to calls, speak on his behalf, know how he thinks, ingest context, and enable junior levels to leverage his expertise and context immediately.
How quickly private markets transform is uncertain. Private markets have been immune to standardization because of unstructured data, which AI should fix. It's unclear if regulatory or market forces will force additional standardization accelerating transparency and liquidity. It's also unclear how much alpha will remain in human relationships. Micro cap M&A with small business owners wanting to trust successors through handshake may not be fully automated, but sponsor-owned businesses, secondaries, private credit, and GP/LP secondaries likely will be.
Gabe's fundraising style is super direct and transparent about concerns and goals. You need to be very headstrong on the end state. It's about reps and relationships - people who lead Series C or D are those met at seed, A, B, C who passed every time but gathered more data. Avoid fake it till you make it - you need bravado and confidence even when not fully confident, but you don't need to pretend to be something you're not. Believe that 5% likelihood of becoming a hundred billion dollar business is likely, delineate with clear roadmap and strategy how it's possible, and have confidence that if those things play out, you will be.
Gabe is paranoid about everything that can go wrong, feeling every day like you're on the knife's edge of a thousand things collapsing. Business building is a game of compounding momentum. He's scared the momentum will stop, hit a roadblock, go off course, and need to revitalize momentum. It's clear how hard it is to build a machine that gathers momentum, so constant preparation for stumbling blocks that halt velocity is essential.
Businesses that solve these problems will become black holes for talent, capital, and brand, able to siphon resources to execute. AI is an amazing tailwind but the execution bar is higher than ever. Everyone is getting pulled into the big leagues with their welcome to the NFL moment. You need to move faster, be stronger, more resilient and agile than ever expected. Having the right team is more important than ever, and the right team is extremely expensive. If you can't figure out strategy to become a black hole for talent and capital quickly, you're at risk of being roadkill of a lab or company that can.
Gabe benefited enormously from parents who were enormously kind, generous, and selfless, showing up differently for his mother versus father. His mother's version of kindness was that no matter what he did, he was amazing and smart and could do no wrong, even though growing up that was absolutely not the case. She instilled confidence to believe in himself even during low trajectories when everything seemed to go sideways and after 40 nos in a row.
The speaker reflects on the contrasting parenting styles of his mother and father and how each contributed to his development. His mother provided unwavering emotional support, treating him as though he was the smartest person on earth regardless of any failures or academic struggles. This irrational confidence stemming from undying love proved valuable despite being technically unwarranted.
His father operated from a completely different approach, maintaining high standards and discipline. When the speaker made mistakes or failed to capitalize on opportunities, his father would become visibly frustrated but channeled this into a methodical process of understanding the root causes. He would sit down and analyze every detail to help improve performance while maintaining his own principles and definitions of excellence.
The conversation concludes with acknowledgment of the business being built and its ongoing improvements. The host expresses appreciation for the guest's participation in the discussion.
Several sponsor messages follow, highlighting how small advantages compound over time in both investing and company operations. RAMP positions spending systems as capital allocation strategies that improve data quality, decision-making, and long-term economics. Vanta scales compliance automation and provides unified security and risk management as businesses grow. Work OS enables AI and software companies like OpenAI, Cursor, and Perplexity to achieve enterprise readiness rapidly. Ridgeline serves as a technology partner for asset management firms, supporting 5x scaling through faster growth and improved operational capabilities.
Keep Invest Like The Best in your library
Save the videos and channels worth coming back to, and find them again in one place.
