Week 6: Notice 06: Legal Discovery / Paralegal Review
Expert work's last clean notice before the break — and the apprenticeship the pitch never counts
29 March 2027
Last notice before the break, and the last one in the clean, confident, early-arc register: a real efficiency gain, a rollback plan that’s genuinely just inconvenient rather than impossible. Read it that way on purpose, so the shift starting week 7 is visible as a shift and not just “the notices got longer.”
Outline#
- the apprenticeship argument specifically: why a task can be worth automating and worth keeping, for two different reasons, at once
- what “a cost that arrives in about five years” means for how a firm should weigh this migration today
- what to watch for as the course crosses the mid-semester break
The pitch
Document review in discovery is reading thousands of pages to find the hundred that matter, tagged for relevance and privilege. A model trained on the matter's own key terms reviews the entire document set overnight, flags every plausibly relevant document with a defensible audit trail, and never needs a second associate brought on to hit the production deadline.
Migration path
The full document set is ingested and scored for relevance and privilege against matter-specific criteria; a much smaller supervising team reviews only the model's flagged set and any documents it scored as uncertain.
Breaking changes
The unsatirized part. What the migration actually costs, plainly.
- Economic
- First- and second-year associate work has long included discovery review specifically because it's how junior lawyers learn a practice area's documents, vocabulary, and fact patterns before they're trusted with judgment calls. Automating the review removes both the billable hours that funded junior associate salaries and the apprenticeship those hours were quietly providing, and firms have not yet built a replacement for the second one.
- Functional
- A document's relevance sometimes turns on context the model wasn't given --- an internal joke that reads as a threat out of context, a term that means something different in this specific deal than in the training data. A reviewing associate builds a working theory of the case while reading, and re-flags earlier documents when that theory changes; a scoring pass doesn't revise its own earlier judgments when it learns something new later in the same set.
Rollback plan
A firm can staff a review the old way for any matter that calls for it, and the tools don't force anyone's hand. But the junior associates who would have learned a practice area through years of review work are the same people a firm would need to supervise the model's output, so under-staffing that apprenticeship now is a cost that arrives in about five years, not immediately.
Teaching team
- Idris Fenn — tutor