Nearly every AmLaw and mid-market firm now has an agentic AI pilot running somewhere in the building. Very few have made it out of the pilot. Everest Group's latest research — and our own field data across dozens of engagements — points to the same pattern: enthusiasm is high, but the operating model, data foundation and governance that turn a pilot into a production capability are usually missing, and no amount of additional pilots will fix that.
Why most pilots stall
The pilots that never scale share three symptoms with remarkable consistency: they are owned by innovation teams instead of practice groups, they run on ungoverned data, and they lack a clear economic thesis. Without a partner in the practice, agents never touch real matters. Without governed data, security and risk block the rollout at the exact moment it needs to expand. Without an economic case, the pilot quietly dies the first time a budget is reviewed by someone who does not personally love AI.
What 'scale' actually requires
Scaling agentic AI is less an AI problem and more an operating-model problem. It requires named practice-group sponsors with real authority, matter-level KPIs measured monthly, a retrieval layer that respects ethical walls and matter scope, a model-routing layer that can move between providers as capability and cost evolve, and a human-in-the-loop pattern that is fully auditable. Firms that put this scaffolding in place move from a handful of pilots to dozens of agents in production within two quarters. Firms that don't stay stuck.
Security and confidentiality are non-negotiable
The fastest way to kill an agentic AI program at a law firm is a leak across matters or clients. Even the perception of a leak is enough to freeze the program for a year. Production-grade agents run on hardened Microsoft 365 or Google Workspace tenants, with role-based access, DLP, tenant-scoped retrieval, per-matter access boundaries and full audit trails. Anything less is a headline waiting to happen — and every managing partner reading this knows exactly which competitor they do not want to become.
The change-management underlayer
The technical work of scaling agentic AI is often easier than the human work. Attorneys need to trust the tools, understand their limits, get proper attribution for using them well, and see their supervisors modeling the same behavior. Firms that pair rollouts with real enablement, office-hours, practice-group champions and named executive sponsorship see adoption in weeks. Firms that assume attorneys will discover the tool on their own see shelfware in months.
A pragmatic path forward
Start with two or three high-friction workflows — intake, conflicts, discovery review, first-pass drafting, matter budgeting narratives — with an explicit economic target for each. Instrument them from day one. Report to the executive committee monthly. Then expand deliberately, one practice group at a time, one workflow at a time. This is how firms move from 'we have a pilot' to 'AI is part of how we practice' — and it is the only pattern we have seen work consistently at scale.
The two-year outlook
By 2028, firms that have not scaled agentic AI beyond a small handful of enthusiasts will be structurally disadvantaged on matter economics, talent retention and client wins. The evidence is already showing up in RFP responses and rate conversations. The window to be an early mover is closing, but it has not closed — and the firms moving deliberately in the next four quarters will still capture a real advantage.
If your agentic AI program is stuck in pilot mode, we can help you build the operating model and guardrails to move it into production without triggering the risk conversations that killed the last attempt. Talk to our AI practice — or subscribe below for our monthly legal-AI briefing.
Contact us
