I’ve spent more than 15 years staring at screens trying to figure out what a market is about to do before it does it. And I’ll confess something few analysts admit out loud: a good chunk of that work was always mechanical —reading a balance sheet, cross-checking numbers, building a model. The hard part, the part that actually protects someone’s capital, was never that.
The Job Already Taken From Analysts
Tomorrow, before the opening bell, JPMorgan and Goldman Sachs report third-quarter earnings. What almost nobody mentions is that both banks are already production customers of Anthropic’s financial agents, launched in May with Jamie Dimon on stage alongside Dario Amodei, alongside Citigroup, Citadel, and Carlyle, among others, used to draft pitchbooks, build models, and close the books (Fortune). Dimon has said he built a research dashboard with Claude Code in about twenty minutes. And Magnetar Capital, the hedge fund firm with $18 billion under management, is going a step further: Bloomberg reported it’s preparing a fund where hundreds of AI agents, not human analysts, will hunt for ideas, analyze companies, and generate investment recommendations (Bloomberg Law). The final buy-or-sell call, for now, stays human.
What Nobody Says Out Loud
The risk isn’t that AI does a junior analyst’s job poorly. It does it reasonably well, and it gets better every month. The real risk is more uncomfortable. When the recommendation arrives already packaged, clear, and well-written by a model, the human temptation is to accept it, not question it. Judgment —the part of the job that actually protects someone’s capital— is exercised by questioning, not approving. Questioning something that sounds as reasonable as what a model trained on two decades of data produces takes more discipline, not less, than questioning the mediocre work of a tired human analyst on a Friday at six p.m. I watched this happen with risk models before 2008: nobody questioned them because the math looked flawless, until it wasn’t. Nobody builds the muscle of questioning by accident; it atrophies if you don’t use it on purpose.
The Opportunity That’s Actually Yours
Here’s the opportunity, and it’s real: in a market where the mechanical part of analysis —reading, summarizing, modeling— gets automated, value concentrates increasingly in whoever knows how to ask the question the model didn’t ask itself. If you work in financial analysis, in treasury, or you’re a student anywhere —including Colombia— deciding what to study with an eye toward a first analyst job at a bank or a multinational, the winning bet isn’t competing with AI on processing speed —no human wins that race anymore. It’s building the habit of skeptical judgment: asking the model to show its assumptions, hunting for the scenario it didn’t consider, and having the technical authority to say “this doesn’t add up” when the rest of the room already accepted the easy answer.
The Question That Matters Before Markets Open
Tomorrow, when JPMorgan and Goldman release their numbers, part of the analysis surrounding them —inside the banks and outside, among those covering the stock— will already have passed through an AI model before it reaches a human eye. That’s not alarming; it’s simply the new starting point, and sooner or later it will reach your industry too, whatever it is —we already saw it with the agents now trading retail accounts on their own. The question I’ll leave you with isn’t whether AI will take your job of analyzing. It’s whether, when the model’s answer arrives polished and convincing, you’ll still have the judgment —and the value— to check it before signing it as your own.
If you want to follow this conversation daily, today’s Daily Briefing goes deeper into what’s coming this week with the banks and the Fed.