AI Doom, Veeva, Research and Writing

Summary

The hosts open with a software investing disclaimer and frame the biggest macro issue as AI-driven disruption risk to SaaS, especially uncertainty about what AI can do over a 5-10 year horizon. They distinguish between two separate questions: whether AI will keep improving, and whether it will merely raise productivity or actually replace whole software layers. A cited bearish AI report is treated as useful only if it is grounded in real business dynamics. The conversation then shifts to Veeva, where they argue the company’s clinical-trial software has a strong moat because it sits inside highly regulated, audit-heavy workflows that end in FDA submission packages. They also discuss Veeva’s CRM history, its conflict with IQVIA over life-sciences data, and the migration off Salesforce toward Veeva Vault. The episode closes with a long exchange on writing: how to develop taste, why AI prose feels soulless, and why research should be done like a judge weighing evidence rather than a lawyer defending a case.

Notes

• The episode starts with an investment disclaimer and ownership disclosure.
• Both speakers say they have exposure to software names, especially Constellation Software and Veeva.
• The macro setup is a software selloff driven by AI disruption fears.
• The hard part is not near-term valuation compression but estimating AI capability five or ten years out.
• SaaS valuation is implicitly a bet on terminal value when you buy long-duration multiples.
• They separate AI progress into two questions: continued capability gains versus substitution risk.
• The framework is whether AI becomes a productivity layer or a replacement layer.
• The Citrini-style bearish report is criticized for not being grounded in underlying business mechanics.
• The complaint is not about bearishness itself, but about weak causal grounding.
• Veeva was founded in 2007 as life-sciences CRM built on Salesforce force.com.
• Its original CRM stack depended on external life-sciences rep data.
• IQVIA One Key is described as the industry-standard source for physician and hospital-affiliation data.
• Veeva built Open Data around 2012-2013 as its own version of One Key.
• IQVIA later blocked uploads of One Key data into Veeva software, triggering litigation.
• The two companies were stuck in a decade-long commercial standstill before a data-use agreement was reached.
• Veeva began migrating CRM off Salesforce and onto its own Vault architecture about two to two and a half years ago.
• CRM is now a mature part of the business rather than the main growth engine.
• CRM once represented most of Veeva revenue but has shrunk to a much smaller share.
• Veeva used to have 18 of the top 20 biopharma CRM customers and may keep 14 of 20.
• Biopharma sales forces have already been shrinking for years, which could pressure CRM demand further.
• Clinical trials are modeled as a production line whose output is an FDA submission package.
• The workflow includes chained regulatory steps, not a loose collection of tasks.
• The speakers emphasize that each step creates new records, approvals, and dependencies.
• That dependency chain is a reason sponsors and CROs standardize on Veeva rather than stitching point solutions together.
• Systems of record in regulated domains are hard to replace.
• Buyers in regulated software also look at what sophisticated peers are using.
• The AI question for Veeva is whether agents sit outside the stack or get absorbed by the incumbent.
• Clinical trial software is treated as a self-contained process with most value likely captured by the incumbent suite.
• If customers can build a better agent on another platform, that would be disastrous for the incumbent.
• The default assumption should not be that the incumbent loses the AI layer.
• Veeva is discussed as trading around 20x LTM EBITDA and roughly 30x earnings or mid-20s forward earnings.
• The speakers think some software names are merely reasonable, not obviously cheap or expensive.
• A core research rule is to act like a judge, not a lawyer.
• They warn that company-defense writing can become PR-like.
• The writing advice is to read broadly, including fiction, to avoid cookie-cutter finance prose.
• The speaker sees AI-generated prose as increasingly common and often soulless.
• AI writing quality is portrayed as stubbornly stuck despite broad model progress.
• AI is useful for critique and editing, but not for taking over the first draft.
• AI has a humor problem.
• OpenAI people reportedly tried and failed to improve writing quality and remove em dashes.