One thing I've noticed: the learning curve for AI in finance isn't linear.
The fundamentals are remarkably accessible today. Free courses cover model behavior, prompting, and workflow integration. With enough time and effort, almost anyone can learn the basics.
But somewhere along that curve, the slope changes.
Put two investment professionals in front of the same model, the same document, and the same basic training. One gets a generic summary. The other gets a specific, actionable read: the covenant package that actually constrains the borrower, the baskets most likely to matter for this capital structure, and the three questions the credit committee is going to ask.
The difference usually isn't the model. Beyond a certain point, it isn't even the prompt.
It's everything that happens before the prompt: how knowledge is organized, what context the model is given, and how the workflow is structured around what the analyst is actually trying to accomplish. Each decision seems small. Together, they determine whether the output is generic or genuinely useful.
The investment professional who gets the better output isn't better at using AI. They know the domain. They know which questions matter. And they're working with a system that consistently gives the model the right context.
Most practitioners are near the beginning of this journey. Starting is the right move. The biggest leap doesn't come from learning one more prompting technique. It comes from combining domain expertise with the right system behind the prompt. That's when the output starts to look fundamentally different.