A first-year associate used to earn their salary the hard way: eighty-hour weeks buried in case digests, statute annotations, and memos nobody would ever read twice. That work is thinning out fast, and firms ignoring the shift are running up a bill they can't see yet, a class of midlevels who can operate an AI tool but can't spot the argument it missed. The training pipeline was built on grunt work. Pull the grunt work out and something has to fill the space.
A handful of comfortable assumptions have taken hold in partner meetings and practice group calls around this shift. Most of them are wrong. Here are the ones worth putting down before the next associate class starts.
Myth: AI Case Summaries Are Just a Faster Version of the Same Junior Task
An AI-generated case summary and a first-year memo are different artifacts that fail in different ways. A junior associate who reads a case wrong fails in predictable ways: a missed footnote, a misread holding, too much weight on the headnotes. An AI tool fails in unfamiliar ones.
It can pull up the right case and describe a holding that isn't in it. It can be confident about a jurisdiction it never checked. That's why coverage of coverage of Law.co's AI-generated case summaries case summaries and the broader wave of citation-backed research tools matters: the summary is only useful if the underlying architecture forces every claim back to a real, retrievable source. Firms evaluating these platforms should test them the way opposing counsel would. Pick cases where the holding is narrow, and see whether the tool honors the narrowness.
Myth: If the Grunt Work Goes Away, Training Takes Care of Itself
The old apprenticeship was clumsy, but it worked because it was repetitive. Read enough motions and you start to hear when a brief is off-key. Take that volume away and you have to replace what it taught, on purpose. A Bloomberg Law analysis found that the foundational work once handed to junior associates, legal research and internal memo drafting, is increasingly being handled by AI, with juniors shifting toward reviewing and refining the output.
Review is a skill. Nobody is born with it. If a first-year's job is now to catch what the model got wrong, the firm has to teach them what a good catch looks like, and give them the reps to build the instinct.
Myth: The Firm Can Wait to See Which Tool Wins
Waiting has a cost that never shows up on an invoice. Every month a firm defers a serious rollout is a month its associates aren't building the review muscle their peers at other firms are building. It's also a month clients are being sold, by someone, on faster turnarounds at lower rates.
An Artificial Lawyer piece made the point plainly: as automation absorbs routine associate work like discovery and due diligence, forward-thinking firms have to move training toward higher-value skills, tech fluency, and validation methods for AI outputs. Firms sitting this out aren't preserving quality; they're handing the decision to the market.
Myth: Clients Won't Notice the Change Under the Hood
General counsel notice. They notice when a memo lands the next morning that used to take a week. They notice when the same associate name shows up on three matters instead of one. They notice when the block-billed research entry gets shorter and the analysis entry gets longer.
Sophisticated buyers are already asking pointed questions about how research is produced, how citations are verified, and who signs off on the output. Firms that can answer with a concrete workflow, private infrastructure, human approval gates, audit trails on every summary, hold their pricing. Firms that mumble through the answer are going to see their rates squeezed by clients who assume the savings should be theirs.
What to Ask Before the Next Associate Class Arrives
A short set of questions worth putting to your practice group leaders before September:
- Written expectations. Do first-years have a written description of which tasks they draft themselves, which to draft with AI, and which to review only?
- Verification standard. Is there a firm-wide standard for how an AI-generated summary or statute analysis gets verified before it goes into a filing?
- Feedback loop. When the tool gets something wrong, where does that finding go, and does anyone act on it?
- Supervision model. Who makes sure a second-year has read enough raw opinions to know when a summary is off?
None of these require a technology budget to answer. They require a partner to decide what a trained associate should look like in five years, and to work backward from there. Firms doing that work now will have midlevels who can think. Firms that don't will have midlevels who can only click.
