What this means for practitioners
Where the boundary sits and how I'd actually use these skills.
Going back even two years ago, demonstrating the power of these tools to anyone in the industry would have left them mind-blown. These skills are still very impressive and can short cut a lot of quickfire or work you want to build on top of. However, the consensual nature of the output, the potential for errors and the need for human judgement are all unresolved issues.
What worked
The power of these tools is in the speed. For updating models on earnings days, for writing quick reaction to lots of newsflow (eg a salestrader's morning note) for parsing a lot of background information on companies - these frameworks work well.
There is also power in the ability to maintain up-to-date coverage on numerous key issues for every stock, particularly for active managers running multiple portfolios. The ability to triage a universe of opportunities and point investors in the right direction for their process is a big plus.
What didn’t
The trust in the information, particularly after watching the intiation skill fall over in my first run is a cause for concern. I also believe there is a natural pull and training data bias towards a more consensual answer. Superior returns lie in doing something different, correctly, away from consensus.
As written elsewhere on this website, I believe outsourcing the decision making and judgement to tools like this is ill-advised. Knowing the mechnism for arriving at the recommendation answer is realy important, as that is then possible to debate. But I am still of the view that humans win in making decisions in complex, uncertainty environments at this stage.
Where it fits in a real workflow
Personally, I would adjust the skills and tools to fit the process I wanted to follow. Getting the same answer everyone else receives when using these tools is not the recipe for success. However, it may be useful for understanding consensus and furthermore the advantage of using these tools may lie in the ability to gather and collate diverse sources of information quickly. It's this key underlying skill which I'd be most keen to utilise as a practitioner.
Where this leaves it
An AI equity analyst is something people have been building towards for a while. Claude's version does achieve something very impressive. But now everyone has access to this same tool, the question then becomes 'how can I use this tool effectively and better to generate superior returns?'
Chapter 7 of 7