An AI Legal Assistant Doesn't Replace Your Practice Management Platform
Anthropic launched Claude for Legal last week. Twelve practice-area plugins, MCP connectors into Docusign, Box, and Thomson Reuters, and the Justice Technology Association signed on as a launch partner. The legal tech world treated it as a category event — and it is one — but it also surfaced a question I keep hearing from small-firm partners:
"If I have an AI legal assistant, do I still need a practice management platform?"
The answer is yes. The reason matters more than the answer.
Two different categories pretending to be the same one
An AI legal assistant is a research and drafting tool. Ask it about a statute. Hand it a document and ask it to summarize. Get a first pass at a clause. It's an extension of your thinking, faster than your memory, and very useful inside a single task.
A practice management platform is the system of record for everything beyond that single task. It holds your matters. It holds your trust accounts. It holds your time entries, billing, invoice approvals, conflict checks, ethical walls, communications threaded to the matters they belong to, and your client list. It's the place a firm operates from.
Those are not the same product; they solve different problems. And the marketing language over the next 18 months will deliberately blur the difference, because both categories want the budget line.
A small test: where does the answer live?
The cleanest way to see which category a tool belongs to is to ask, for any given question, where the answer is supposed to live afterward.
If I ask a research assistant, "What's the standard non-compete enforceability test in Washington?" the answer lives in my head, or in a memo I write next. The assistant doesn't need to store the question or the answer for me to use it later — once the answer is mine, the assistant's job is done.
If I ask my practice management platform, "What's the status of the Henderson matter?" the answer has to live in the system. It has to know the matter exists. It has to know which documents are attached to it, which time entries have been logged against it, what the trust balance is, who has access, who doesn't, when the last client communication was, and whether anything is overdue.
The first question — research — can be answered by any sufficiently smart model with the right reference material. The second question — operations — can only be answered by the platform that holds the matter.
If you try to use Claude for Legal (or any general AI legal assistant) to answer the Henderson question, the assistant will give you a confident answer assembled from whatever data it can see — but it doesn't have the underlying state of your firm. It has whatever you've pasted into the chat or connected via MCP. That's still just a research surface, not a record system.
Anthropic's own architecture says this
This is the part that gets overlooked in the launch coverage. The Claude for Legal MCP connectors go into existing systems — DocuSign for signatures, Box for documents, Thomson Reuters for research. Anthropic isn't claiming Claude becomes your system of record. Anthropic is saying Claude can read from and act on your existing systems of record.
Which means: the architecture assumes you still have those systems. The pricing, the productivity story, and the entire pitch presume that the firm has somewhere to anchor matters, store documents, track time, and run billing. Claude is the assistant; your stack is still your stack.
For a firm with a clean stack — modern practice management, modern document management, modern billing — Claude for Legal is an incredible accelerator. For a firm running on six fragmented tools with no unified matter layer, plugging in a general-purpose AI assistant means the assistant inherits the same fragmentation. You get a smarter drafter sitting on top of the same broken plumbing.
What a practice management platform does that an assistant can't
Some things the assistant category, by design, isn't trying to do:
Trust accounting. Frozen Retainer, Standard, Evergreen — IOLTA balances tracked against matter activity, with the audit trail your state bar will ask for. An AI assistant can describe trust accounting rules. It can't hold the money.
Passive time tracking. Time entries are generated from your actual activity across the platform, not from end-of-day memory reconstruction. An assistant can summarize a meeting. It can't watch every action across every matter and turn that into accurate billable time.
Ethical walls. Configurable barriers between teams, matters, and clients, enforced at the data-access layer. An assistant can be told what it shouldn't see. A platform refuses to show it in the first place.
Invoice approval workflows. Multi-level review, write-offs, discounts, credit notes, all tied to matters and clients. Assistants don't have a workflow engine.
Multi-tenancy and RBAC. Every query enforces org-level tenant isolation, with role-based access controls verified on every route. This is not a sexy feature; it's the part that makes the platform safe to use for a regulated profession.
None of this is interesting until it breaks. When it breaks — a misrouted trust withdrawal, an associate seeing a conflicted matter, a missing time entry on a six-figure case — it's the firm's problem, not the assistant's.
The right pattern is integration, not substitution
The smart move for a small firm is not to pick between an AI legal assistant and a practice management platform. It's to choose a practice management platform that has AI features built in where the matters and time and trust accounts already live. Verification, document classification, privilege detection, key-date extraction — these belong inside the system of record, because the value of running AI in a legal practice is bounded by how clean and complete the data it's running against is.
A general-purpose AI assistant connected to a fragmented six-tool stack will produce confident answers from incomplete data. A practice management platform with AI features built into its matter architecture will produce confident answers from a unified record. The first is a research surface. The second is operations.
Both can be useful. Don't confuse them. When you're shopping, ask the vendor a direct question: Are you a research and drafting tool that connects to my systems, or are you the system?
If they say both, they're one of them and bolting on the other. Figure out which.