AI and the Infinite To-Do List

We hosted a series of Portfolio Manager dinners in July where we asked how they are using AI in their process. The range of answers was wide. The divergence turned out to be the most useful part of the evening. Some have systems with connectors into their datasets with agent workflows. Others were in the earlier stages and asking good questions about where to start.

A major topic was Skills. A Skill is a structured prompt you build up over time, so when you notice an area for improvement, you write the correction in, and it holds for the next run. One manager described running a standing devil's advocate against every thesis before an analyst’s idea pitch. Another looks for external sources that support or contradict the thesis. Another has a daily summary agent that gives them the daily update for every position in the portfolio. This is something their trader used to have to build every morning. Each Skill is a piece of a firm's judgment and process codified so that it permanently part of the investment process.

Another topic was architecture. The managers furthest along treat the model as a hub that houses their skills and the data required for AI to do its best work. Examples of data they wanted their hub to have was emails, research (stored in files or RMS), market data, sell side research, portfolio data, EDGAR, and financial models. Research, internal and external, seemed to be the stickiest problem. Either because their internal research was locked inside of a 3rd party system and the sell side didn’t always provide a way to connect. One PM at a multi-billion dollar manager called their hub the “castle walls.” They spend their time pulling the data and skills inside the “walls” while being agnostic to the underlying models (LLMs) so that they could choose the best for the task.

A thoughtful stretch of the conversations was about how analysts learn. When AI builds a finished first pass, a young analyst can reach the answer without building the instinct that comes from working through it. There was no silver-bullet answer for building instinct. The only agreement was that it is the responsibility of the PM is to make sure the analyst understands what AI has delivered.

I asked each table what they thought AI would do to hiring. The initial response was that nobody is going to hire anyone, headcount holds roughly where it is, and you no longer need six people in operations to run operations.

I took the other side and said that we’ll be hiring more. The clearest case I have is our own developers. Someone who used to finish five real projects in a year now finishes something closer to twenty five. The cost for the developer is the same, so the return on that person went up 5x. We direct investment dollars toward the highest return for the same reason you size a position that way. We have an “infinite to do list” of things worth building has not gotten any shorter. So I expect to hire more talent proficient with these tools rather than fewer.

One final thought. AI is great at answers. Thoughtful and insightful questions are the scarce input. Creativity in how we use AI is the skill that separates people. The people who are good at asking thoughtful questions are more valuable than ever before. That is where I would put the effort, and it is where we are spending most of ours right now.

‍