Scuttleslops, OpenAI valuation, Ryan Specialty

Summary

The hosts discuss Scuttle Slops, a mostly AI-generated paid summary service, as a case study in where AI helps most: narrow, repetitive tasks like earnings and conference recaps. They spend time on disclosure norms, arguing that undisclosed AI-generated writing can damage trust even if the output is good. The conversation then shifts to OpenAI, where one speaker models the company needing roughly $700 billion of 2035 revenue, about 55% gross margin, and roughly 22% operating margin to justify a ~10% IRR, concluding the bar is extremely high. The final section examines Ryan Specialty and specialty insurance, explaining how wholesale brokerage and delegated underwriting work, why the market is concentrated, and why AI may erode some expertise advantages but not the relationship-heavy parts of the business. The broad takeaway is that the episode is less about headline AI hype than about practical constraints: disclosure, economics, margins, and which human roles remain durable.

Notes

• Scuttle Slops is presented as a mostly AI-generated summary product with light human analysis.
• The host says he still spot-checks the output and makes editorial choices.
• The product is meant to satisfy demand for higher-velocity updates on names already covered.
• Long-form writing is reserved for consequential events that change a company's long-term positioning.
• AI is framed as better for narrow tasks like earnings summaries than for strategic analysis.
• The host says AI still struggles with long narratives and prioritization.
• The speakers expect the human/AI content boundary to keep blurring.
• Passing off copied AI text as original work is described as bad form.
• AI-assisted brainstorming is treated as a gray area because humans and models can co-develop ideas.
• Speaker B argues AI interactions will shape beliefs over time and blur authorship at the cognitive level.
• The host says AI-generated slop triggers a strong negative aesthetic reaction.
• He says A16Z Substack posts appeared to be mostly AI-written.
• The host says hiding AI use would preserve secrecy but lose goodwill.
• Speaker B notes that some outlets disclose AI assistance, using Skift as an example.
• The discussion mentions Pangram as an AI detector, but not as definitive proof.
• The conversation moves into OpenAI and Anthropic as the focal frontier-lab names.
• Reported OpenAI revenue and losses are used as the starting point for the model.
• The model says OpenAI would need about $700 billion of revenue in 2035 to generate roughly 10% IRR.
• The implied 2035 operating margin is about 22%, or about $155 billion of operating profit.
• Gross margin is modeled to expand from 43% in 2025 to 55% in 2035.
• Training cost is excluded from COGS in the model; COGS is described as inference only.
• R&D is treated as the main drag, including model training runs.
• 2026 R&D is expected to exceed $40 billion, implying R&D near 100% of revenue.
• The model assumes R&D peaks around $127 billion in 2030 and flattens near $140 billion by 2035.
• Sales and marketing falls to about 8% of revenue by 2035, while G&A falls below 5%.
• Speaker B concludes that the valuation bar is so high that even AGI upside may not make the investment worthwhile.
• The host says frontier labs sit in the 'too hard' pile because he lacks intuition for the revenue and profit assumptions.
• He contrasts a brute-force AGI bet with a more traditional moat-based software path.
• Speaker B says Anthropic appears to be adding around $10 billion of run-rate revenue per month.
• Speaker B questions whether frontier models will remain broadly accessible through APIs over the long run.
• He suggests the only durable customers for the most advanced models may be the US government and the labs themselves.
• Ryan Specialty is introduced as a broker in specialty insurance.
• Standard lines are for well-understood risks; excess and surplus lines cover unusual or hard-to-underwrite risks.
• Examples of specialty risks include experimental malpractice, cannabis dispensary property, and haunted-house liability.
• The speakers argue climate change, AI liability, and autonomous vehicles create new specialty-insurance demand.
• A more different 2035 is framed as a tailwind for specialty brokers.
• Specialty market share is cyclical and tied to hard markets.
• Most specialty share gains since 2003 are said to have happened in the last four to five years.
• Ryan Specialty is described as a wholesale broker and an MGU/MGA with an extra intermediation layer.
• Wholesale brokers exist because retail brokers are too small and niche-specific to build direct carrier relationships.
• Speaker A says the relationship-based rationale for wholesalers is more durable than the expertise rationale under AI.
• Carrier fees to wholesalers are estimated at 15% to 20% of premiums, with about half passed through to retail brokers.
• The wholesale brokerage market is described as an oligopoly led by AmWins, Ryan, and CRC.
• Ryan’s brokerage business is said to take no underwriting risk and just place risk.
• Ryan's delegated-underwriting role is described as an insurer handing underwriting authority to the firm.
• Examples of delegated underwriting products include post-harvest agriculture property, amusement parks, and hazmat trucking commercial auto.
• The delegated-underwriting market is described as fragmented because carriers outsource niche risks they do not want to underwrite internally.
• Ryan says it avoids conflicts by not owning retail brokers.
• Insurance distribution is described as a mix of cooperation, competition, and Chinese walls.
• Insurance brokers are said to be trading like software stocks because of an AI overhang and a softer cycle.
• AI may make some risks more legible, but specialty placements are not like online auto quotes.
• Specialty brokerage still depends on tower structuring, carrier appetite, and tacit knowledge.
• Ryan is said to pay about 60% of commissions and fees as compensation, versus about 40% at Brown & Brown’s wholesale and delegated underwriting businesses.
• The compensation gap is framed as both a margin gap and a retention risk.
• Brown & Brown is cited as the biggest victim of Howden's poaching, with about 300 employees taken.
• Insurance brokers are said to be down 50% to 60% without a recession, which the speakers treat as unusually severe.
• The brokerage roll-up model is compared to Danaher and Thermo Fisher.
• Large brokerage acquisitions are described as costing about 15x to 16x EBITDA, while smaller targets can be bought at 7x to 8x.
• Constellation Software is cited as a better roll-up model because it can buy many smaller targets at lower multiples.