The latest legal AI developments show the market moving beyond standalone prompts. Harvey has launched Command Center for peer-based usage visibility and partnered with DeepJudge, while Lavern has launched as an open-source agentic legal system with specialist agents and workflows. At the same time, commentary on legal translation risk is a useful reminder: outputs that look right can still be wrong if the workflow does not test legal meaning, context, and consequences.
Usage measurement is becoming part of the AI workflow
Artificial Lawyer reports that Harvey has launched Command Center, a peer-based window into how customers are using different Harvey features. According to the article, firms and in-house teams can allow anonymous usage data to be shared with peers, with visibility by feature and practice area.
For firm leaders, the important point is not the dashboard itself. It is the operating habit behind it. If a firm is investing in AI, leadership needs to know where adoption is happening, which practice areas are moving faster or slower, and whether usage is translating into better internal workflows.
That matters for custom AI as well. A bespoke research, drafting, intake, or knowledge-management workflow should not be treated as a one-time software launch. It needs measurement: who uses it, where it helps, where it stalls, and which teams need process redesign rather than another tool.
Agentic systems raise the bar for workflow design
Artificial Lawyer also reports that Lavern has formally launched and offers 67 specialist legal agents across eight workflows. The article describes Lavern as a figurative agentic "law firm," not a law firm in the current sense, and frames the broader question of what work is handled by agents versus by human judgment.
That framing is commercially useful for law and accounting firms. The right question is not whether an agent can replace a professional. The better question is which repeatable steps can be structured, routed, checked, and escalated so that professionals spend more time on judgment.
For a firm building custom AI, agentic workflows should start with bounded tasks: gather documents, classify facts, summarize a record, compare clauses, prepare a first-pass issue list, or route a matter to the right reviewer. The workflow should make the handoff to a human explicit, especially where legal, tax, or client-risk judgment is required.
Correct language is not the same as correct professional advice
A separate Artificial Lawyer article on legal terminology warns that legal AI can produce outputs that look correct linguistically and terminologically while still being wrong in substance. The summary highlights cross-border work, where two concepts may align at the level of terminology but differ in purpose, scope, conditions of application, and legal consequences.
This is a critical design lesson for any professional-services AI workflow. A system that drafts a clean answer is not necessarily a system that has resolved the professional issue. In legal, tax, audit, and advisory work, the workflow has to test assumptions, jurisdictional context, source fit, and the consequences of applying a concept.
That is where custom AI should differ from a generic chatbot. The firm can define required checks, approved sources, reviewer roles, exception paths, and evaluation sets for the specific work it performs. The goal is not more fluent output. The goal is a workflow that helps the firm reach a safer and more useful work product.
Professional teams still need technology fluency
In an interview with Artificial Lawyer, Intel's Joy Sherrod said she does not agree that the next generation of in-house lawyers needs to be technologists first and lawyers second. She said companies hire in-house attorneys for specialized knowledge and advice, while also noting that attorneys need to become conversant in AI tools to become more efficient and spend less time on lower-value work.
That balance applies across professional services. Lawyers, accountants, and advisors do not need to become software engineers. But they do need enough fluency to identify where AI belongs in the workflow, where it should not be trusted, and how its output should be reviewed.
For firm owners, this means AI adoption is partly a training and management problem. The firms that benefit will be the ones that pair domain experts with well-scoped workflows, clear review obligations, and practical measurement of usage and outcomes.
- Treat AI adoption as an operating system, not a tool rollout: measure usage by team, workflow, and matter type where possible.
- Use agents for bounded, repeatable steps, then design clear escalation points for professional judgment.
- Do not equate polished AI output with correctness; build checks for context, source fit, and legal or technical consequences.
- Train professionals to be AI-conversant enough to supervise workflows, not to become technologists first.
Sources watched
- Harvey Launches Command Center, Partners With DeepJudge (Artificial Lawyer)
- Legal Innovators California Interview: Joy Sherrod, Intel (Artificial Lawyer)
- When Legal Terminology is Correct But the Answer is Still Wrong (Artificial Lawyer)
- Lavern the Agentic 'Law Firm' Has Arrived (Artificial Lawyer)
