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

Episode Summary

Executive Summary: The discussion argues that AI is now a practical, durable tool for investment research—not AGI—and that its value depends on how thoughtfully it is integrated into existing workflows. Brett Carrin and David Plon explain that recent model improvements, especially deep research and finance-tuned systems, have shifted AI from novelty to competitive necessity, while warning that conviction, skepticism, and human judgment remain essential.

Main Topics: AI’s shift from hype to practical utility (Priority: 5/5): Both guests say the last six months marked a real inflection point: models are materially better, finance-specific tooling has matured, and more asset managers now see AI as a competitive advantage rather than a toy. Workflow design and process decomposition (Priority: 5/5): They stress that AI should not replace the investment process wholesale. Instead, analysts should map their research workflow into discrete steps and use AI to augment specific tasks where it adds leverage. General-purpose models vs. specialized tools (Priority: 4/5): ChatGPT/Claude/Gemini are useful for broad research, summarization, and getting up to speed, while tools like NotebookLM and Portrait are better for grounded workflows, private-document analysis, thesis monitoring, and idea generation. Conviction, skepticism, and the analyst’s edge (Priority: 5/5): The speakers argue that AI can accelerate work but cannot substitute for judgment. Good prompts, domain expertise, and skepticism are necessary to avoid consensus answers, hallucinations, and weak conclusions. Adoption patterns across the buy side (Priority: 4/5): Contrary to assumptions, smaller, resource-constrained funds are often faster adopters because AI meaningfully expands bandwidth. Larger funds face more compliance, data-privacy, and workflow-integration barriers. Limits of current AI in investing (Priority: 5/5): The strongest warning is that LLMs remain weak on math, dates, precision, and structured financial modeling. They are useful for qualitative work, but not yet reliable for complex quantitative modeling or end-to-end trading decisions. Boot camp and education around implementation (Priority: 3/5): Brett explains the upcoming AI boot camp will focus on evidence-based use cases, guest practitioners, case studies, and prompt templates to help analysts adopt AI without getting lost in hype.

Key Arguments: AI is only as good as the questions and process behind it; using it well requires decomposing the investment workflow into discrete steps. Recent model improvements, especially deep research capabilities, have made AI genuinely useful for ideation, research synthesis, and thesis monitoring. General-purpose LLMs are best for broad retrieval and summarization, while specialized tools are needed for grounded, document-based, and repeatable workflows. A robust AI system in finance is largely a software engineering problem, not just a data science problem. Smaller funds adopt AI faster because it gives them institutional-grade bandwidth without requiring large teams. Larger funds may use AI more as an alpha-generation tool than an efficiency tool, but compliance and privacy constraints slow adoption. AI cannot yet be trusted to handle high-precision math, dates, adjusted financial data, or fully automated modeling without human review. The biggest risk is outsourcing reasoning; AI should accelerate or enrich conviction-building, not replace it. Good prompting mirrors good analyst management: clear, context-rich instructions with access to the right data and explicit expectations. The industry is moving toward agentic overlays that monitor theses, alert on KPIs, and integrate across Excel, notes, RMS systems, and news. Early adopters who build workflow and data plumbing now will be best positioned to benefit from future model improvements. Current outputs often reflect consensus views unless users ask highly specific, differentiated questions grounded in primary sources.

Data Points: Time since last discussion: Six months - Brett says the biggest shift in AI adoption and capability has occurred since the prior podcast appearance in the spring. Initial ChatGPT release referenced: November 2022 - Brett describes first experimenting with GPT-3.5 when it came out. AI deployment level in workflow: 25%–30% - Brett estimates AI is now embedded in roughly a quarter to a third of his research workflow, up from about 2% six months earlier. Prior workflow use: 2% - Brett contrasts current AI usage with his estimate from six months earlier. Lower-cost AI stack: $20 per month - Brett notes a strong AI stack can be built with a low-cost consumer subscription. Higher-end AI stack: $30,000 per year - Brett contrasts consumer tools with expensive institutional software and copilots. Prompt length: 9 words - Brett cites a statistic that the average prompt in foundation models is only nine words, highlighting weak prompting habits. Finance boot camp scale: 15 guest speakers - Brett says the AI boot camp will interview many practitioners and builders to provide evidence-based guidance. Adoption target threshold: One-third of workflow - Brett argues outsourcing more than one-third of research motion to AI is probably too much at present. AI adoption horizon: 2026–2027 - David says the most advanced users will be best positioned to exploit future model releases like GPT-6 and beyond. AI invention year referenced: 1956 - Brett notes AI has been one of the most hyped technologies since its inception.

Pivotal Quotes: "AI will make smart people smarter and dumb people dumber." — Opening framing / Max: Sets the central thesis of the episode: AI amplifies capability rather than replacing it. "This technology is only as useful as you direct it today." — Opening framing / Max: Introduces the idea that workflow quality and prompting discipline determine AI’s value. "If more than one-third of your investment research motion is outsourced to AI, that's probably too much." — Brett Carrin: Brett’s rule-of-thumb for preserving judgment and conviction in public markets.

Implications: AI is becoming a durable layer in investment research, but success will depend on workflow design, data plumbing, and user sophistication. The winners will pair AI leverage with skepticism, domain expertise, and human judgment.