Workflow Ownership Replaces Model Moat

Legal AI is no longer won by having the smartest model, it is won by owning the lawyer's workflow and the systems around it. Harvey dropped its custom legal model after frontier models beat it on BigLaw Bench, then rebuilt around multi model agents, implementation teams, and integrations into research databases and document systems. That shifts the moat from model training to deployment, trust, and daily usage inside firms.

• Harvey's product now looks less like a single legal brain and more like an orchestration layer. One model analyzes documents, another handles research, another drafts, and ex lawyers help firms set it up so attorneys actually use it enough to renew.
  1 sacra 2 sacra
• That change lowers the advantage of early vertical model builders and opens the market to fast followers like Legora, which skipped fine tuning and went straight to market with frontier models for international law firms.
  2 sacra 4 sacra
• The next competitive fight is over workflow ownership and data access. Clio is pairing practice management with vLex's legal corpus, while Harvey Vault and DeepJudge are competing to sit on top of firm documents, permissions, and matter history where grounded answers become useful work product.
  2 sacra 3 sacra

From here, legal AI should consolidate around companies that control the full path from source material to finished work. The winners will be the products that can pull the right precedent, respect access controls, draft inside existing tools, and fit naturally into how firms bill, review, and deliver legal work.