Legal Workflow, Not Models, Wins

This marks Harvey’s shift from selling smarter legal answers to selling a better legal work system. Once OpenAI, Anthropic, Google, and xAI models could match or beat Harvey’s own tuned model on legal reasoning benchmarks, the hard part stopped being generating a decent memo or clause and started being pulling the right documents, routing work through firm systems, and getting lawyers to use the product every day inside live matters.1 2 3 4

• Inside large firms, Harvey and Legora are increasingly close on core model capability, so adoption is being decided by client pull, integrations like iManage, training, and license management. Firms often buy small seat counts, hot swap licenses across teams, and deploy by practice group rather than firmwide.1 2
• That opens room for narrower products to win specific workflows. Spellbook focuses on contract drafting and redlining, DeepJudge focuses on search across internal firm documents, and Clio uses its vLex acquisition to bundle legal research corpus with workflow software that Harvey still has to stitch together through outside systems.3 4
• Harvey’s response has been to go multi model and put more weight on implementation. Its product now matters less as a standalone model and more as an orchestrator that chains models, firm knowledge, and review steps into a usable workflow, supported by ex lawyers in customer success who drive rollout and utilization.2 3 5

The next battleground is ownership of the legal workflow surface, not ownership of the smartest model. The winners are likely to be the products that become the default place where lawyers search internal knowledge, review contracts, draft first versions, and move matters forward with auditability and deep system integration.1 2 4