Legal AI Fragmenting into Wedges
Glean for law
This shift means legal AI is no longer one market with one winner, it is breaking into narrow products that own a specific screen, document flow, or customer type. Once frontier models made raw legal reasoning easier to buy, advantage moved to workflow fit. Spellbook lives inside Word for contract redlines, DeepJudge sits on top of a firm's internal document systems, and incumbents like Clio are buying data and workflow assets to stitch those wedges back into a broader stack.
-
Contract work is the clearest wedge because the ROI is immediate. Reviewing an NDA, MSA, or DPA is repetitive, time sensitive, and already happens in Word and email, so tools like Spellbook can cut turnaround from 1 to 2 days to minutes without asking firms to change their whole stack.
-
Search is a separate wedge because legal answers depend on the right source set and permissions. DeepJudge focuses on searching a firm's own work product in systems like iManage, where access controls and ethical walls matter as much as model quality, while Harvey Vault uses a more manual curated document approach.
-
Incumbents are responding by vertically integrating around existing distribution. Clio reached about $300M ARR by June 2025 serving firms across practice management, then bought vLex for $1B to add a global legal research corpus, aiming to combine documents, matters, billing, and research in one system instead of leaving AI startups to own the front end.
The next phase is a land grab for the legal system of record. Standalone wedges will keep winning fast in narrow jobs, but the biggest platforms will be the ones that connect matter data, internal documents, outside law, billing, and agent workflows into one trusted workspace. That is where legal AI stops being a copilot and starts running the operating layer of legal work.