What We Learned Backing Casetext: AI Research in the Law Firm Workflow

Eighteen months after leading Casetext's Series A, we have learned things about AI adoption in law firms that we did not know when we wrote the check. The company's growth in the period has validated parts of our thesis and revised others. The delta is worth examining — not to revise the investment, which we would make again — but to refine our understanding of how AI spreads in institutional legal settings.

What we expected: attorneys would adopt AI research tools incrementally, starting with associates at large firms and spreading upward as efficiency gains became visible. We expected adoption to be faster at firms that had already invested in legal technology infrastructure and slower at firms with legacy practice management systems.

What actually happened: adoption was fastest at small and mid-size firms, not large ones. The reason was straightforward — large firms have research departments, associates, and formal library resources. For a senior partner at a 12-attorney litigation firm handling multiple active cases simultaneously, AI legal research was not a productivity improvement. It was an operational necessity. The tool was expanding what one attorney could reasonably manage, not accelerating what a research team was already doing.

The second thing we got right, more clearly than we expected: the compliance dimension of the product matters enormously to law firm buyers. The concerns were not about hallucination in the general AI sense — attorneys are sophisticated enough to know that any research tool requires verification. The concerns were about professional responsibility: if an attorney submits a brief citing a case found through AI research, what is their duty to verify? How does the firm's professional responsibility insurance treat AI-assisted work product? These questions are being answered case by case, state bar by state bar, but they shape procurement decisions at every firm we spoke to.

The third learning: integration with existing workflow tools is table-stakes. Attorneys who have to leave their document management system to use a research tool use it less. The firms with the highest adoption had the deepest workflow integrations. This seems obvious in retrospect but was underweighted in our initial thesis.