> ## Content Index
> Fetch the complete content index at: https://www.dueprocessor.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# The Missing Rung: AI Is Doing the Junior's Job. Who Trains the Next Partner?
- URL: https://www.dueprocessor.com/the-missing-rung-ai-is-doing-the-juniors-job-who-trains-the-next-partner/
- Published: 2026-11-11T14:11:05.000Z
- Updated: 2026-11-11T14:11:05.000Z
- Author: Dave Kimelberg

AI can now do much of the work that used to fill a first-year associate's week, and it does it faster and cheaper. What nobody has worked out is how lawyers will learn judgment once that work is gone.

## The first-year's job, automated

Think about the tasks a new associate has traditionally been handed. Summarize the deposition. Review ten thousand documents for various issues. Build the chronology. Pull every change-of-control clause from the data room. Review thousands of documents in a diligence data room and provide a summary. Write the research memo on the limitations period.

The ABA's ethics committee listed the uses of generative AI in [Formal Opinion 512](https://www.americanbar.org/groups/judicial/resources/appellate-issues/2026-fall/apprenticeship-problem-age-of-ai?ref=dueprocessor.com) in 2024: research, document review, compliance work, drafting. As one appellate commentator put it this fall, in ordinary law-firm parlance, that is a junior-associate task list.

These are exactly the jobs today's tools do well. A first pass that took a week now takes an hour. A partner who once waited for a draft can now ask for one and get it in the afternoon.

## The efficiency case is real

The savings are no longer hypothetical. In Wolters Kluwer's [2026 Future Ready Lawyer Survey](https://wolterskluwer.com/en/expert-insights/legal-ai-adoption-time-savings-revenue-growth?ref=dueprocessor.com), 92% of legal professionals said they use at least one AI tool every day. Sixty-two percent reported saving 6% to 20% of their week, and about four in five said the tools meet their expectations.

The gains land hardest on document review, contract review, and first drafts, the base of the work pyramid (and the business pyramind). Clients have noticed. In-house counsel now press outside firms on their AI plans and expect the efficiency to show up in the bill. Some are balking outright at paying for deposition summaries and document review ([Hartford Business Journal](https://hartfordbusiness.com/article/ai-driven-efficiencies-force-law-accounting-firms-to-rethink-the-billable-hour/?ref=dueprocessor.com)).

For a client, this is simply good news. Paying a junior associate $300 to $400 an hour (and higher for Big Law firms) to grind through a data room was always a hard sell, but accepted. AI gives them a reason the firm can't argue with.

## Training was always a byproduct

The law firm pyramid is built on leverage. Firms hire large associate classes, bill their hours at a markup, and keep the spread between those rates and associate pay. A few partners sit on top of many associates, and the hours at the bottom fund the whole structure.

Training rode along for free. As Three Crowns CEO Hugh Carlson and Stanford's Megan Ma put it in [Bloomberg Law](https://news.bloomberglaw.com/us-law-week/ai-flight-simulators-will-replace-grunt-work-in-lawyer-training?ref=dueprocessor.com), firms billed for many of the hours associates spent learning on the job, so the apprenticeship paid for itself. Nobody had a reason to build anything else.

AI breaks that arrangement from both ends. Clients won't pay for hours a tool could have saved. And firms that adopt the tools need fewer junior hours for the same matter. The work that once trained associates, and paid for their training, is shrinking at the same time.

So far, the shift shows up in support staff more than in associate classes. Baker McKenzie cut 600 to 1,000 business services roles in February 2026, and Clifford Chance made similar cuts a year earlier, citing AI ([Slaw](https://www.slaw.ca/2026/07/31/the-hidden-economics-of-the-vanishing-apprenticeship/?ref=dueprocessor.com)). U.S. legal employment, meanwhile, sat at a ten-year high. The associate pipeline hasn't narrowed ***yet***. The work feeding it has.

## The question nobody has answered

The drudge work was never valuable only because clients paid for it. It was how lawyers learned to see and understand.

Reviewing thousands of documents teaches you which one matters. Drafting discovery teaches you that loose words become expensive disputes. A bad first memo, marked up by a partner, teaches you what a good one looks like and why. Judgment is pattern recognition, and patterns come from repetition.

AI hands the junior lawyer the answer without the patterns underneath it. The memo is clean, the cites are real, the summary is crisp. But the associate who produced it may not know why it is right, or notice when it is wrong. It upskills without building the skill.

That matters because supervising AI requires the very judgment AI is removing the chance to build. Someone has to catch the hallucinated case, the missing contrary authority, the clause that is off-market. Today that someone is a senior lawyer who learned the hard way. In ten years, who will it be?

There is early evidence the worry is real. A Harvard Business School field study of Boston Consulting Group consultants found more than a quarter became "self-automators," handing whole workflows to AI without picking up new skills ([Bloomberg Law](https://news.bloomberglaw.com/us-law-week/ai-flight-simulators-will-replace-grunt-work-in-lawyer-training?ref=dueprocessor.com)). In a Russell Reynolds survey of more than 1,750 global leaders, 54% worried over-reliance on AI is stunting critical thinking and judgment.

The risk isn't that firms stop hiring juniors tomorrow. It's that they keep hiring them, give them nothing formative to do, and find out in five years that their midlevels can't be trusted with a matter.

## What firms are trying

Firms and commentators have started to experiment, but none of these approaches has a track record yet. Each tries to rebuild deliberately what the old model produced by accident.

| Approach                        | Who is trying it                                                                                                                 | What it is meant to build                                                 | What we don't know yet                                              |
| ------------------------------- | -------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------- | ------------------------------------------------------------------- |
| Billable credit for AI learning | Ropes & Gray lets first-years put up to 20% of a 1,900-hour target, nearly 400 hours, toward AI training and experiments         | Fluency with the tools, and room to rethink how a task gets done          | Whether tool fluency turns into legal judgment                      |
| AI boot camps                   | Latham & Watkins put its 400-plus first-years through a mandatory two-day AI Academy                                             | Shared standards for when and how to use AI                               | Whether two days changes daily habits                               |
| Simulation                      | Three Crowns built Atelier, an AI cross-examination simulator with a feedback agent                                              | Specific instincts through practice and feedback, like a flight simulator | The authors concede it can't teach the profession's unwritten rules |
| Audit, don't draft              | Recommended by ABA: juniors check AI output against sources before seniors see it                                                | The verification skill that supervising AI requires                       | Whether you can learn to check work you have never done yourself    |
| Deliberately slow work          | Slaw columnist Susan Van Dyke suggests some tasks be done without AI                                                             | Judgment built through friction                                           | Who pays for the slower hours                                       |
| New junior roles                | Tom Martin of the Thomson Reuters Institute proposes rotations such as AI compliance, legal data analysis and knowledge curation | Technical and operational skills firms now need                           | Whether these roles lead to partnership-level judgment              |
| Earlier client contact          | Stewart McKelvey's Paul Saunders urges bringing juniors into client meetings and real-time mentoring                             | Client sense and the unwritten rules                                      | Whether partners have time, and whether clients accept it           |

The common thread is cost. Every one of these asks the firm to fund training directly, instead of hiding it in billable hours. Clients have made clear they won't pay for it as "training." They might pay for verification, audit trails, and lower error rates, if firms can show the link.

## The rung we're removing

AI is good at junior work, and clients are right to want the savings. Nobody should preserve make-work so new lawyers can suffer the way their elders did.

But the bottom rung of the ladder did two jobs. It produced work, and it produced lawyers. We have found a faster way to do the first job. We have not found a replacement for the second, and the firms that assume one will appear on its own are likely to discover the gap when their first AI-native class reaches the midlevel years.

So the question for every firm, law school and client is not whether to use AI. It is who will pay to build judgment now that the billable hour no longer does it for free, and what that training will actually look like.

I don't think anyone knows yet. If your firm is trying something that works, or something that didn't, I'd like to hear about it.