The Edge Has Shifted | Matt Reustle on How the Best Investors Use AI

Episode Summary

Executive Summary: The discussion argues that AI is now materially useful for investors not by creating obvious alpha, but by dramatically improving research, monitoring, screening, writing, and workflow efficiency. Matt Russell explains how deep research, agentic tools, and better prompting let investors customize outputs, surface missed information, kill bad ideas earlier, and run leaner teams—while still requiring human judgment, source verification, and experimentation.

Main Topics: From novelty to professional utility (Priority: 5/5): The conversation traces AI’s evolution from early ChatGPT novelty to genuinely useful deep research and agentic workflows that can support investor research and decision-making. LLMs vs. agents in investing workflows (Priority: 5/5): Russell distinguishes single-shot LLM responses from agentic systems that reason across multiple sources, ask follow-up questions implicitly, and produce analyst-like outputs. Workflow automation and monitoring (Priority: 5/5): AI is presented as a way to monitor portfolio names, customer commentary, sector signals, and recurring thematic issues that investors previously missed due to time constraints. Prompting, customization, and mental models (Priority: 4/5): A major theme is that output quality depends heavily on how well an investor encodes their own heuristics, objectives, audience, and desired format into prompts and systems. Trust, verification, and hallucination control (Priority: 4/5): Russell stresses source checking, tighter prompts, and constrained reasoning to reduce hallucination risk, while noting that human analysts also make errors. Tool stack and experimentation (Priority: 4/5): He describes a multi-tool setup—Claude for writing, ChatGPT for general use, Gemini/ChatGPT deep research, DIA for browser-based interaction—and emphasizes iterative testing. Industry impact and future adoption (Priority: 4/5): The likely near-term effect is leaner teams and better productivity rather than automated investment decisions; long-term, customization and agentic assistants may reshape how research is consumed.

Key Arguments: AI’s biggest current value for investors is leverage: faster research, broader coverage, and better monitoring rather than a direct, obvious alpha signal. Early LLMs felt like social or search tools, but deep research and agentic models became meaningfully useful for professional investing in late 2024/early 2025. Agentic workflows matter because they can pull from more sources, reason through layers of a question, and produce outputs closer to what a junior analyst would deliver. Real-time monitoring is especially powerful because investors can now track recurring themes, customer commentary, and sector-specific signals without manually combing through every transcript. AI helps investors kill bad ideas earlier by screening for deal-breakers such as accounting issues, incentive misalignment, or business-model fragility. Prompt quality is a decisive variable; many shortcomings attributed to the model are actually prompt-design failures or under-specified objectives. Investors should verify sources and maintain human oversight because investing remains a judgment process, unlike a purely mechanical yes/no task. The most effective adoption path is bottom-up experimentation by individual users, then sharing use cases internally, rather than forcing a top-down build that may never get used. Off-the-shelf investor-specific tools may be more effective than custom builds unless the firm has highly specific workflows and substantial data to encode. AI may compress the gap between specialists and strong generalists by accelerating information gathering and synthesis, but it does not eliminate domain expertise or decision-making skill.

Data Points: Timing of major inflection in AI utility: Late 2024 to early 2025 - Russell says deep research models and premium tiers marked the point when AI became genuinely useful for professional investors. Months ahead: David Plan is described as 6-9 months ahead; Russell claims he is about 4-6 months ahead of the host - Used to frame relative adoption and knowledge curves in AI usage. Adoption cadence: Daily, weekly, monthly - Examples of recurring monitoring schedules that can be set up for alerts and digests. Prompt testing frequency: Anytime a new model is released - Best users compare standardized prompts against prior model outputs whenever models update. Team cadence example: Friday 30-minute stand-up - Russell cites this as a useful forum for sharing AI use cases inside an organization. Typical range of model output control: Less vs. more flexibility in reasoning - He notes users can tune models to be more creative and exploratory or more constrained to reduce hallucination risk.

Pivotal Quotes: "I still don't think it's solved this really obvious alpha-generating solution." — Matt Russell: Russell’s core caveat that AI currently boosts productivity more than it directly creates repeatable investment alpha. "The best people that I know at using AI have very specific systems, which anytime a new model is released, they'll have their prompts, which are detailed, and they'll see what this new model does in terms of output and be able to compare that to the previous model." — Matt Russell: He explains how advanced users standardize prompts and run controlled comparisons across model releases. "You can put in a monitor for any type of interesting commentary around logistics coming from... and set it up so that it's done on a recurring basis and sending you information that otherwise you just would have never seen." — Matt Russell: Illustrates the practical monitoring and alerting use case for investors tracking sector-specific signals.

Implications: Investors who adopt AI well will likely work faster, cover more ground, and miss fewer signals. The edge shifts from access to information toward better prompting, judgment, and workflow design—especially for lean teams and customized research processes.