The Race to Be Finance's AI Backbone: Why Deep Workflows Trump Frontier Models
Key points
Key takeaways from an Invest Like the Best interview with Rogo co-founder and CEO Gabriel Stengel (September 2026):
Why the labs won't win finance. Stengel argues that winning financial services is not about frontier intelligence — OpenAI and Anthropic focus on the model layer, while the durable value sits in proprietary plumbing: firms' data sets, systems of record, compliance and audit workflows. He advises building products "perpendicular" to what the labs want to build, and likens their chasing a vertical's depth to "stopping on the side of the road to pick up a penny" on the path from $100bn to $1tn in revenue.
A 10-year shift to systems of record. He poses a thought experiment for incumbents: if in 10 years the world's best investment firms and banks hold 90% of their enterprise value in software, data and systems rather than in people, you would start extracting what lives in the latent minds of your best people into a system you own and operate. He cites Jane Street taking roughly 15 years to build the dominant quant franchise as the benchmark for how long the AI transition will take.
Dealmakers, not public-equity desks, are the core market first. Rogo's primary users sit on either side of a transaction — preparing data rooms, answering due-diligence questionnaires and coordinating deals through closing. Half of Rogo's surface area is integration with firms' CRM, portfolio-monitoring and LP-distribution systems, and private markets appeal precisely because they are still run by humans and largely unautomated, whereas public markets are already automated.
Product eras track model eras. Rogo's capability steps map to frontier models: o1 Pro first made the product reliable enough as a search tool, and with Opus 4.5 in late 2025 and early 2026 the models became capable of essentially any junior analyst or banker task given the right instructions and context. Stengel says the first-mover disadvantage of building ahead of the models flips into an advantage once they arrive — but only if you bet on the right end state.
Auditability matters more than raw accuracy. As Rogo moves from co-pilot to autopilot, clients need to debug every output — the data inputs, the assumptions and the full lineage of an AI-generated decision. Compliance adds to the plumbing burden, from handling material non-public information to the prospect that a Delaware court could make AI inputs and research discoverable.
Per-seat pricing today, outcome pricing tomorrow. Rogo prices per seat, positioned in the same category as Bloomberg, FactSet and Capital IQ, and so requires a human go-to-market machine rather than the pure token-consumption model of a "token broker". Stengel expects every AI business to move through usage-based and then outcome-based pricing, and says he would rather charge per good investment idea or per deliverable than assign dollars to tokens.
Skills that still matter. Durable value sits in what he calls "move 37" thinking — gathering proprietary data, building relationship graphs of experts and feeding a model with inputs no one else has — plus judgment that AI has yet to prove it can replace. He also flags the productivity shift for senior bankers: a managing director can now email a marked-up deck to Rogo's AI analyst and get it back in 20 minutes instead of two days.
Lessons from a tough Series A and the "company brain". Stengel recounts being turned down by around 40 investors while raising the Series A — including Sequoia, Kleiner Perkins, Benchmark and, after a long courtship, Thrive Capital — before Keith Rabois backed the company as "basically Harvey for finance". Internally, every conversation is recorded and filtered into Rogo's "company brain", nicknamed Shrek, which powers onboarding, enablement and knowledge compounding across the firm.
Read more: YouTube · Invest Like the Best
