Harvey Shifts To Workflow Moat

Harvey’s moat moved up the stack from owning a better legal model to owning the legal workflow. Once OpenAI, Anthropic, Google, and xAI models could all do high quality legal reasoning, a custom model stopped mattering as much as the system that decides which model to use, pulls the right documents, checks citations, and fits into how firms actually draft, review, and research across matters.

• The trigger was benchmark performance. By mid 2025, frontier reasoning models were beating Harvey’s fine tuned legal model on Harvey’s own BigLaw Bench, so Harvey shifted to multi step workflows that can use one model for document analysis, another for research, and another for drafting.
  1 sacra 2 sacra
• That shift also changed where Harvey had to invest. Instead of spending to keep a proprietary model ahead, it put more weight on workflow setup, tool orchestration, and ex lawyer customer success staff who help firms turn AI into repeatable daily work instead of occasional prompting.
  1 sacra 2 sacra 5 sacra
• It also narrowed Harvey’s path to owning legal research. The June 18, 2025 LexisNexis alliance brought citation backed answers into Harvey, but through an integration with Lexis content and AI. At the same time, incumbents were deepening their own stacks through acquisitions like Thomson Reuters buying Casetext and Clio buying vLex.
  3 sacra 6 lexisnexis 7 thomsonreuters 8 clio

From here, legal AI competition is likely to center on who owns the most trusted end to end workflow, not who has the flashiest model. The winners will combine model routing, document system access, research integrations, auditability, and firm specific context into a product lawyers can rely on every day for high volume work.