PhD-for-Hire: Deep Research and the End of the Junior Analyst's Moat

This month I watched an AI produce a competitive-landscape briefing — sourced, structured, footnoted — of the kind I used to assign to sharp junior analysts with a one-week deadline. It took twenty-six minutes.

Deep Research is the productization of something profound: not answers, but research — the multi-hour grind of gathering, reading, reconciling, and synthesizing that constitutes the actual workday of millions of knowledge workers. Armies of us built careers on exactly this work. It was the apprenticeship layer of the professions — the years where you learned the craft by doing the grind. The grind is now a product feature.

The same month, Andrej Karpathy coined "vibe coding" — describing software built by conversing with a model rather than typing syntax — and the phrase went from joke to job description in about six weeks. And in Paris, the international AI summit circuit completed its pivot: two years ago these gatherings were about existential safety; this one was about "opportunity," investment, and national competitiveness. Labor transition, again, appeared mostly as a communiqué clause.

The apprenticeship problem

Here is what keeps me up at night — not the senior expert being replaced, but the ladder to becoming one being pulled up. If the machine does the junior work, where do juniors come from? Professions that solve this — deliberately funding human apprenticeship even when the machine is cheaper — will still exist in twenty years with deep benches. Professions that don't will discover they traded their future for one budget cycle's margin.

Actionable Steps

Firms: treat junior roles as R&D for your talent pipeline, not as cost centers the agent just beat on price. Redesign them around supervising machine research and owning small decisions early. Educators: the essay-as-assessment is dead; the defensible skills are framing questions, judging sources, and standing behind conclusions. Individuals: learn to direct research agents this quarter — the person who can commission and quality-check five parallel investigations is worth more than the person who can do one by hand.

The moat was never doing the reading. It was knowing what to read for. Guard that.