A second-year associate opens her laptop and the memo is already written, with the intake summary attached, the research pull footnoted, and a draft contract sitting in the matter folder flagged for her review.
Six months ago, writing those documents was her week. Today it's a notification.
That shift is working through law firms, accounting practices, consultancies, and software teams, and it has scrambled the oldest training model knowledge work has. The Chicago Journal's coverage of the junior job question in the age of agentic AI walks through what the redesign looks like in practice. An agent now does the first draft in three minutes, so the real question is whether anyone still has a way to learn the trade.
The Monday Morning Memo Was Doing Two Jobs
The tedious work a junior used to grind through was billable to the client and educational for the junior at the same time. One activity, two outputs. Firms priced the first and absorbed the second as a cost of making people. A commentary in Fortune recently put it bluntly: entry-level roles were secret apprenticeships subsidized by firms, and AI has cut the subsidy.
When an agent writes the memo, you still get the deliverable, and you lose the second output without anyone noticing because nothing on the invoice changes. That's the part that catches firms off guard a year or two in, when they look up and realize the mid-levels aren't ready to take over the work the seniors are tired of doing.
The Hiring Numbers Are Already Moving
The clearest signal that this isn't theoretical is in the labor data. Stanford's Digital Economy Lab has tracked a measurable pullback in entry-level hiring concentrated in roles where AI substitutes directly for the junior's tasks, with early-career workers in highly AI-exposed occupations falling well behind their less-exposed peers. The pattern isn't layoffs. It's a hiring slowdown, with firms swapping the apprentice headcount for a seat license and a review workflow.
That math works in the short run and breaks in the long run. The senior who's reviewing agent output today was a junior five years ago, drafting the memos nobody now writes. The pipeline that produced her isn't producing her replacement.
Rebuild the Apprenticeship Inside the Review Loop
The on-ramp has to move. If the first draft is the agent's job, then the training ground becomes the review: reading the agent's output, catching what's wrong, deciding what to keep, and understanding why. That's a real skill, and it's teachable, but only when juniors are given the time, context, and feedback to develop it rather than being handed a merge button and a quota.
A few things change when you take that seriously:
- Junior review time becomes billable work. If a mid-level reviews the agent's draft and a junior reviews it alongside her and marks it up first, that second pair of eyes is the apprenticeship, and it needs a line on the matter plan rather than being shoved into after-hours shadow work.
- Agents expose their reasoning. A black-box answer teaches nothing. An agent that cites which clause it pulled from, which precedent it leaned on, and which assumption it made gives the junior something to agree or disagree with.
- Seniors mark up the review, not just the draft. The feedback loop that used to run on the first draft now has to run on the junior's edits of the agent's first draft. Same muscle, one level up.
This Only Works When the Agent Lives Inside Your Own Systems
A generic chatbot bolted onto the side of the firm can't do any of this. It doesn't know your templates, your matter history, your house style, or the specific way your partners want a settlement memo structured. It produces fluent output that a junior has no particular reason to trust or distrust, which is the worst possible training environment.
Agents become teaching tools only when they're wired into the firm's own document stores, matter management, code repositories, or client records, and only when the review workflow is part of the same system. Most buyers underestimate that piece. You're not procuring a model.
You're commissioning the pipeline the model runs inside: the retrieval layer that grounds its answers in your work, the review queue that routes drafts to the right humans, the audit trail that lets a senior see what the junior changed and why. Firms that want to keep training people are increasingly building that pipeline themselves, often with a custom software partner, because the off-the-shelf tools stop at the draft and leave the apprenticeship problem on the floor.
The Firms That Figure This Out Keep Their Bench
Five years from now, two kinds of practices will exist. One kind will have ridden the productivity curve, shipped more work with fewer juniors, and discovered their mid-level bench is thin and their seniors are exhausted. The other kind will have spent the same five years rebuilding the apprenticeship inside the agent-assisted workflow by paying for review time, demanding transparent agents, protecting un-automated hours, and owning the systems all of it runs on.
The deliverable looks the same from the outside. The bench doesn't. And the bench is what you sell.
