How AI Tools Are Rapidly Disrupting the Investment Industry | Brett Caughran & David Plon

Episode Summary

Executive Summary: The conversation argues that AI is now a practical, durable tool for fundamental investing—not AGI hype—but its value depends on process design, data discipline, and user skill. Brett Carrin and David Plon describe a clear inflection in adoption over the last six months, especially for research acceleration, thesis monitoring, and idea generation, while warning that AI still struggles with math, dates, and unstructured trustworthiness. They conclude that AI should augment, not replace, conviction-building workflows.

Main Topics: AI has moved from novelty to practical investing tool (Priority: 5/5): Both guests say the last six months brought a noticeable jump in model quality and institutional adoption, especially after O3 with deep research and broader ecosystem improvements. Workflow decomposition and prompt quality (Priority: 5/5): Successful AI use requires breaking the research process into discrete steps and writing detailed prompts that match specific workflows rather than asking for full thesis generation. Data grounding, hallucination control, and compliance (Priority: 5/5): General-purpose models are useful for broad research, but grounded systems like Notebook LM and finance-tuned copilots matter because public-web answers can hallucinate and internal documents often raise compliance issues. Human judgment remains central to conviction (Priority: 5/5): AI can speed up research and broaden coverage, but it cannot replace skepticism, context, or the reasoning required to build differentiated investment conviction. Adoption patterns differ by fund size and maturity (Priority: 4/5): Smaller, resource-constrained managers may benefit most immediately, while large funds face integration, privacy, and workflow constraints; they are more likely to use AI for alpha enhancement than pure efficiency. The next frontier is agentic, integrated operating systems (Priority: 4/5): The speakers envision AI overlays that monitor theses, alert on KPI inflections, and connect Excel, notes, RMS systems, and news into a smarter research operating system. Boot camp and education as evidence-based adoption (Priority: 3/5): Their AI boot camp is designed to teach practical use cases through case studies, guest managers, and prompt templates, avoiding hype-driven narratives.

Key Arguments: AI is here to stay even without AGI; incremental engineering gains alone make it valuable for high-stakes intellectual work. The best near-term use cases are research acceleration, idea generation, thesis monitoring, summarization, and getting up to speed on a company or sector. You cannot accelerate a broken process; firms must first document their research workflow before deciding where AI adds leverage. General-purpose models are broad retrieval and synthesis tools, but grounded tools reduce hallucination by anchoring to user-uploaded documents and high-quality source data. Prompting works best when it resembles assigning a clear task to a junior analyst with known capabilities and data access. AI should augment conviction-building, not outsource it; otherwise users get consensus views, generic outputs, and lower-quality decisions. Smaller funds and clean-sheet startups are often faster adopters because they lack staff and can redesign workflows from scratch. Large funds face constraints from privacy, counterparty agreements, and existing legacy systems, making adoption slower and more complex. The biggest limitations today are quantitative accuracy, especially math, dates, and modeled financial outputs, where AI is still unreliable. Future value will come from agentic systems that combine research, alerts, and internal data into an integrated operating layer for investors.

Data Points: Time horizon for adoption change: Last 6 months - Brett says institutional AI usage and usefulness inflected materially over the last half-year. AI research take-up assessment: 25–30% - Brett estimates the share of his research motion now augmented by AI, up from about 2% six months earlier. Prior baseline AI usage: 2% - Brett describes his own usage level roughly six months before the current inflection. Maximum AI outsourcing threshold: More than one-third is probably too much - Brett warns that if over a third of research is outsourced to AI, the process may be over-automated. ChatGPT subscription cost: $20 - David notes that even a basic paid subscription can deliver substantial value for getting up to speed and transcript summarization. High-end AI stack cost: $30,000 per year - Brett contrasts low-cost general-purpose tools with more expensive finance-tuned ecosystems. Small-fund AUM example: $30 million AUM - Brett uses this as an archetype of a small manager who may have enough capital to run a fund but not enough to build a full team. Boot camp timing: September - The AI boot camp they are launching is scheduled for September. Foundation model launch reference: November 2022 - Brett references GPT-3.5’s release as his first encounter with the technology. AI inception year: 1956 - Brett notes AI has been hyped since its formal inception in 1956. Alternative data adoption window: 5–7 years - Brett argues early alternative-data alpha windows lasted roughly this long before becoming more embedded in prices. Curriculum size: 15 guest speakers - Brett says the boot camp will include interviews with managers, consultants, and builders using AI in finance. Curriculum structure: 12 modules - Brett describes the analyst academy as being decomposed into a dozen or so discrete workflow modules. Prompt library size: 150 prompts - Brett says they are building a prompt library that maps workflow steps into reusable prompts.

Pivotal Quotes: "AI will make smart people smarter and dumb people dumber." — Unattributed opening framing: The conversation opens with a core thesis about unequal gains from AI adoption. "You can't accelerate a broken process." — Brett Carrin: He explains that firms must define their research workflow before trying to apply AI to it. "Ultimately, this technology is only as useful as you direct it today." — Brett Carrin: He emphasizes that strong prompting, context, and process design determine output quality.

Implications: AI is becoming a core investing skill, not a side experiment. Firms that document workflows, ground their data, and train analysts well will likely gain speed and edge; those that overtrust generic outputs or skip process design risk weaker conviction and bad decisions.