Most everyone in systematic investing is now using the same LLMs, the same data, the same infra. What is edge if the tools are the same? Edge is no longer in the tools; it is the discipline and knowledge in using the tools.

A well-appreciated risk in AI-augmented research workflows right now is variability and hallucination of AI output, which is why we see relatively few systematic shops directly using AI in their models and execution. However, we see dramatic expansion in testing and use of datasets from historically untested sources. All of which is taking place at a time when we know that everyone, everywhere, is using Claude to augment, enhance, and streamline their workflows. Which raises the question, to quote Obi-Wan Kenobi, "Who is more foolish? The fool or the fool who follows him?" If we know not to trust the output of an LLM directly into our models, we should be even more careful to ensure that we understand the workflows and data we are consuming to ensure they are not just a tin-can telephone (or series of them) leading back, ultimately, to Claude, causing subtle model and signal crowding at a previously impossible scale.

The folks who will ultimately be successful are those with structured validation layers around their AI pipelines: source attribution at the point of generation, cross-referenced against primary data and verified human expertise, from recognized sources with multiple checkpoints, as well as idiosyncratic, non-AI-based insight. One-off interactions with so-called "experts," prompted (and perhaps answered) by Claude, will only serve to further bias the system.

A second-order effect worth mentioning is speed. AI compresses research timelines dramatically. If signal validation infrastructure does not keep pace with signal generation infrastructure, you are not performing faster; you are making mistakes faster. The willingness to slow down the model is itself a source of alpha.

Evan Reich is Chief Product Officer and Head of AI at BWG Global, the premier primary research provider for forum, channel, and survey industry intelligence to investment firms and corporations, with a brand-new quantitative data offering, please reach out for information. He previously served in a variety of data leadership roles at hedge funds including Verition Fund Management, BlueMountain Capital Management, Millennium Partners, and SAC Capital. He serves as an instructor for the eCornell AI in Finance certificate program and is a frequent industry speaker on the intersection of AI and data in the investment community.