Two-partner IP litigation shop
Prosecution work bottlenecked on citation checks and prior-art review. We rebuilt evidence as a graph of anchors and turned research into a compounding stock rather than a folder of memos.
Practice
India · 2 partners + 4
Transformation
10 weeks
Track
AI-Native Transformation
Primitives moved
Evidence, Research, Team
The challenge
Every opposition brief required re-verifying citations that had already been verified in a prior matter, because nobody could find the old research quickly. Citation errors had twice reached a hearing before being caught, both times by luck rather than process.
The transformation, phase by phase
Phase 01
Name
We traced twelve recent briefs and mapped every citation back to its source document - most existed only in the drafting associate's memory, not in the file.
Phase 02
Systematise
A standing rule: no assertion ships without a page-and-paragraph anchor. Research answers moved out of scattered memos into one searchable stock, tagged by question and jurisdiction.
Phase 03
Automate
An evidence-checking agent flags any unanchored claim in a draft before it reaches a partner for review; a research agent checks the firm's own stock before generating anything new.
The shift to AI-native
The manifesto's answer to AI hallucination isn't a better model - it's a standing rule that unanchored claims don't ship, applied to humans and agents alike. That rule is what let this six-person shop hand real drafting work to an agent without increasing risk: an AI-drafted paragraph with no anchor now fails the identical check a senior partner's unverified assertion would fail. Two partners with no leverage pyramid to protect could make that rule absolute overnight, in a way a three-hundred-lawyer firm - answerable to a committee on anything touching malpractice exposure - structurally cannot. The firm's research also stopped evaporating: what used to be re-derived from zero on every matter now compounds, which is the quiet advantage the manifesto says teams like this underrate most.
AI-native markers this firm now meets
- AI runs first, on every matter, systematically
- The firm gets measurably better over time
- Technical capability sits inside the firm
Outcome
Citation errors
→ 0 in QA
Prior-art hours
−72%
Repeat questions
answered from stock
Adnah refused half the things I asked for. That was the sign they were serious. Everyone else had said yes to everything.