38% of Legal Operations in Agentic Production — Governance Lags by 12–18 Months

The Legal Stack's Legal AI Agentic Deployment Readiness Report 2026 — surveying 200 legal operations and IT leaders at AmLaw 200 firms, regional firms, and Fortune 1000 in-house departments — found that 38% of respondents have at least one agentic workflow in production (systems taking consequential actions without per-action human approval), 29% are in active pilot, and only 12% have no agentic activity. AmLaw 200 respondents lead at 61% production deployment, Fortune 1000 in-house at 47%. The three proving grounds are court deadline monitoring and docketing (41% of litigation-adjacent respondents with at least a pilot, but only 19% have tested stale-context failure scenarios), NDA routing and execution (63% of in-house CLM users have at least some NDA workflows executing without per-document attorney sign-off, with only 34% having formally documented parameters reviewed in the past 12 months), and outside counsel invoice approval queues (44% of in-house respondents, average autonomous invoice value of $8,400, thresholds up to $25,000). The governance gap is acute: only 11% of all respondents meet all four operational readiness criteria (written authorization policy, tested rollback protocols, stale-context failure testing, and knowledge of which vendor contracts contain agentic liability terms), while 58% self-describe as "agentic ready." The liability structure is similarly fragile: 67% of respondents with agentic vendor contracts report standard SaaS liability caps copied from text-generation tools, and only 19% have contracts with differentiated treatment of agentic-action errors — meaning the organizations bearing the most operational risk from autonomous AI action have the least contractual protection against it.