What Actually Happens When AI Hits a Law Firm

two identical firms · one discovers its new operating model, one designs it on paper · and what that does to the billable hour, the redline, and the leverage pyramid

The Question

A law firm runs on a simple machine: associates bill the hours that pay the partners. AI can now draft the motions, run the research, review the documents. So the real question isn't which tool to buy — it's can a firm change how it works — discovered through the work itself, or designed on paper first?

one firm → two identical copies
One archetypal firm
Aldrich & Vale LLP — AmLaw 100 commercial-litigation firm, ~900 attorneys, ~$3M profit per partner. It runs on the leverage model: the more hours associates bill, the more the equity partners take home. Cloned twice; every run sees the same world.
Same world — same events
  • Market recession
  • Lateral-partner poaching
  • Client AFA / fee mandates
  • Court rule changes
Same matter lifecycle workflow
1
Intake
2
Conflicts
3
Engagement
4
Staffing
5
Research
6
Doc review
7
Drafting
8
Partner review
9
Filing
10
Settlement
11
Billing
12
Collection
A “seam” is the handoff between two people — associate to partner, partner to client — where what one person knows doesn’t fully travel to the next. “Translation debt” is what it costs the firm when something gets lost at that handoff.
Scored at every handoff
risk of things getting lost
tacit vs. codifiable
translation debt
redline rework rate
realization rate
who trusts AI · partners vs. associates
16 quarterly sprints · the only variable is how change happens
Track A

SWT Organic Discovery

Stuart Winter-Tear thesis · implement, observe, then redesign
  • Deploy AI into one bounded workflow — drafting assistance for motion practice
  • No reorganization up front
  • Observe what breaks for 2 sprints
  • Extract learning from what breaks
  • Redesign ONE seam based on evidence
evidence-gated
Track B

Partnership / Consensus

law-native traditional change management
  • Compensation committee debates
  • Equity-partner consensus-building
  • Rainmaker-partner veto
  • A-priori standardization of the workflow's seams
  • A "Target Operating Model" designed before implementation
timeline-gated
where does each approach win? → how tacit the handoff is
Track B
Track A
codifiable  —  template · schematacit  —  expert's head
00.250.50.751
Track B wins — low tacitness

A-priori standardization wins where the seam is codifiable. The seam is designed before it runs.

e.g. legal research · filing
Track A wins — high tacitness

Evidence-built redesign wins where the seam is tacit. The design emerges from what actually breaks.

e.g. partner redlines · settlement negotiation
both tracks run on a hot-swappable tool layer
Descrybe
Case law + primary law engine
what it can do · how accurate · where judgment still rules
Westlaw AI
Legal research + drafting assist
what it can do · how accurate · where judgment still rules
Lexis AI
Research + analysis
what it can do · how accurate · where judgment still rules
CoCounsel
Document + litigation workflow
what it can do · how accurate · where judgment still rules
Harvey
Firm-grade AI platform
what it can do · how accurate · where judgment still rules

The tools are the same in both firms — the question is how each firm puts them to work. And the tool can be swapped out, so the result isn't an artifact of any one vendor.

G1
Surprises us
Unexpected discoveries, not foregone conclusions.
G2
Practitioners recognize it
Face validity with lawyers who live the workflow.
G3
Robust across seeds
Findings hold across random seeds.
G4
Not a straw man
Track B is built as its strongest legitimate form.
G5
Honest counterfactual
Track B can win under the right parameters.

one firm · two tracks · same world · 16 sprints · divergence measured on translation debt · redline rework · realization · trust