I spent 45 minutes building the AI that might replace me

I built an iterative idea-generation tool in under an hour. It works. That's the problem.

I spent less than 45 minutes building a novel AI tool that I think every investor should use…. but, in the process, I’m having an existential crisis about how much longer humans will be relevant to the investing process.

Let me start with the tool: I scheduled a recurring task on Claude Cowork. Every Sunday, it generates five new ideas for me to research that week, along with the reason it recommended each idea and a short summary of the thesis.

Then, throughout the week, whenever I stumble on an idea I like (in an investor letter, on a substack post, etc.), I toss the idea (with a link to the write up or a copy of the PDF) into the project the task lives in. I also include a few lines on what I like about the idea that I’m adding.

Finally, whenever I research an idea that Claude has given me, I give it feedback on the idea (you recommended XYZ and I didn’t like it for this reason; I really liked your recommendation of ABC, etc.).

I’ve been doing this for a few weeks…. and the project has gotten really good. For example, last week its top recommendation was ANAB, which had just done a taxable spinoff with a looming return of capital decision (as well as a bit of ongoing legal drama). If you were playing “Andrew Walker buzzwords bingo”….. well, you’d have just hit the jackpot with that sentence. The other recommendations from the project were pretty solid as well: one recent spinoff, two 13-Ds where the activist pitch focused on an SOTP unlock, and one company that was spun ~a year ago and set to get added to a bunch of indices in the near term.

Importantly, the project has been improving. For example, the first iterations were really centered on super microcap 13-Ds, and over time I’ve trained it that market caps get graded on a loose curve (a $10m company needs just an overwhelmingly insane pitch to get recommended, a $50m company needs a much stronger pitch than a $250m company, and a $2.5B and a $5B company are basically equal).

So that’s the crazy useful AI “tool” that I’ve kind of custom built. IDK, maybe I’m overestimating how novel this idea is. It does seem pretty obvious, but I talk to a lot of investors and I haven’t had a single one tell me about using AI for iterative idea generation like this.

But if I find that tool so useful / novel, why is it causing an existential crisis for me?

Right now, the tool is just helping me find unique ideas that fit my style…. but run that process forward with a few months of training and AI improvement. If I give that system enough feedback, at some point am I even relevant to the process? Couldn’t it identify ideas with my trained “Andrew Walker” filter and eventually decide what ideas to put on and take off?

And remember that I built this model on my own with consumer grade resources. If you’re a multi-billion dollar manager with a nearly unlimited IT budget plus access to actual software developers and a historical backlog of millions of trades over the past few years from all of your teams, how much further could you take my idea? How close are they already to the point I’m worried about hitting (where the model can identify and execute trades without the human)?

As someone who invests for a living, it terrifies me. A few months ago, there’s no chance I could have built a tool like this off the shelf, and AI is improving so fast that I built this little project with less than an hour of prompting. How much further / faster can it go? Are humans going to have jobs as investors anymore? Are we even going to be able to compete with AI in a few months?

Again, I’m terrified… but I think the answer is clear. Humans are going to increasingly need to answer one question: “What exactly am I adding that the AI cannot learn?” The edge continues to move away from simply reading 10-Ks all day and toward finding non-obvious data, asking better questions, and making judgment calls in places where the information is still messy.

P.S. Most of the investors I talk to are still using AI as a senior analyst for initial research, a souped up version of Google, or a way to automate web scrapes. Those are all great use cases, but I think they barely scratch the surface of what these tools can do. If you’re using AI in a genuinely interesting way to become a better investor, I’d love to chat.