Workflow Moat in Legal AI
Scott Stevenson, CEO of Spellbook, on building Cursor for contracts
The real moat in legal AI is owning the workflow, not owning a generic answer box. Once frontier models can all summarize, draft, and answer legal questions reasonably well, the defensible product shifts to tools embedded in the contract job itself, inside Word, inside review flows, and inside proprietary datasets like lease benchmarks that turn raw model output into concrete decisions a lawyer can act on.
-
Spellbook is built around contract drafting and redlining in Microsoft Word, where a lawyer is already marking up an NDA, lease, or MSA. That is different from opening a separate chat window and asking broad legal questions, because the product sits directly in the document and helps finish the task faster.
-
Harvey and Legora have grown by selling broad legal assistants to firms and legal teams, but research across the category shows legal reasoning itself is getting commoditized by frontier models. That pushes value toward packaged workflows, proprietary data, and deployment inside systems of record, rather than chat alone.
-
This pattern has already played out in adjacent AI categories. General purpose tools create awareness and demand at the top of funnel, then specialized products win by doing the last mile work, like comparing contract clauses to market norms, routing approvals, or turning a redline into a structured negotiation workflow.
The next layer of competition is moving from legal chat to contract infrastructure. Products that can ingest large contract volumes, benchmark them against private corpora, and plug into approval, negotiation, and repository systems will be harder to displace than standalone assistants, because they become part of how legal work gets done, not just how lawyers ask questions.