AI Didn't Replace the Junior Analyst. It Changed What the Job Is

The first prediction about AI and white-collar work was the dramatic one: the junior roles disappear. Two years on, that's not quite what happened, and the actual story is more useful to understand. The tasks that filled a junior analyst's week — pulling data, formatting decks, first-draft research — are exactly the ones a competent model now does in seconds.

So the work didn't vanish. It moved up the stack. The valuable thing was never the formatting; it was knowing whether the answer was right, what question to ask next, and how to tell when the confident-sounding output is confidently wrong. AI is excellent at producing a plausible draft and indifferent to whether the draft is true. Someone has to own that gap, and that someone is now expected to be junior.

The training problem nobody fixed

Here's the catch. People used to develop judgment by doing the grunt work — building the model by hand, finding the error the hard way, sitting through the boring parts where intuition quietly forms. Remove the grunt work and you remove the apprenticeship that produced senior people. We've automated the rung of the ladder everyone used to climb.

Firms haven't reckoned with this yet. They've happily taken the productivity, and they're slowly noticing that a junior who never struggled through the manual version is worse at catching the machine's mistakes. You can't supervise an output well if you've never produced it yourself.

What the job becomes

The role that survives is part editor, part skeptic, part translator between a tireless tool and a human decision. It rewards taste and verification over raw output. That's a more interesting job than the one it replaced — but it's also a harder one to start, because it asks for judgment up front instead of letting you earn it.

The organizations that figure this out will deliberately rebuild the apprenticeship — make people do some things the slow way precisely so they learn. The ones that don't will wake up in five years with a generation of analysts who can prompt beautifully and can't tell when the answer is nonsense.