Foundation Models Expand Legal Funnel
Scott Stevenson, CEO of Spellbook, on building Cursor for contracts
Foundation models are expanding the legal AI market faster than they are capturing it. General tools get lawyers to try AI for a quick question or draft, but the durable spend shifts to products embedded in the actual contract workflow, inside Word, with playbooks, track changes, intake, approvals, and system integrations. That is why Claude related publicity lifted Spellbook's demo volume, instead of replacing it, and why chat shaped products face the most direct pressure from model labs.
-
Spellbook sits where legal work already happens. It reviews and redlines agreements in Microsoft Word, applies company playbooks across high volume contracts, and increasingly handles multi document drafting. That last mile product work is far more specific than a general chat interface.
-
The market is splitting by buyer and workflow. Harvey and Legora have centered on broader legal copilots sold top down into firms, while Spellbook has grown bottom up from individual users and now gets about 60% of revenue from corporate in house teams, where faster contract turnaround directly helps the business.
-
This pattern shows up across legal AI more broadly. Contract drafting and review is the clearest wedge because contracts are repetitive, high volume, and easy to benchmark against standards. That makes general AI a strong demo, but specialized products the better operating tool.
Going forward, foundation models will keep widening the funnel, while value concentrates in products that own the workflow, data, and interface around contracts. The winners will look less like a smarter chat box and more like always on contract infrastructure, triaging intake, drafting first passes, and routing work across legal, procurement, and sales.