AI Workflows / First Look
Harvey or Claude or Both? A Practical Routing Rule for Law Firms After Claude's Legal Expansion
Anthropic just wired Claude into the legal stack, including Harvey, Westlaw, and CoCounsel. That changes the buy-versus-build question for firms. Here is a plain routing rule for when to reach for a vertical legal platform and when to reach for the general model.
The either-or just collapsed
For two years the legal AI decision was framed as a fork. A firm could buy a vertical platform built for law, like Harvey, or it could use a general model, like Claude, and bolt legal habits onto it. Most firms picked a lane and defended it. That framing got a lot weaker on May 12, 2026, when Anthropic expanded Claude's legal capabilities with twelve new plugins and connectors that link Claude directly to Thomson Reuters CoCounsel and Westlaw, Harvey, Box, Everlaw, and DocuSign (Reuters, May 12, 2026). At nearly the same time, Harvey rolled out a Command Center for managing enterprise AI adoption.
When the general model can call the vertical platform, and the vertical platform sits inside a managed adoption layer, "which one" is the wrong question. The right one is which tool should do which job.
What each kind of tool is actually good at
A vertical legal platform is built around the things general models historically got wrong in law: grounding answers in real, citable authority, running structured workflows like document review and diligence, and operating inside the guardrails a firm needs. When the task is citation-bound research, a defined review workflow, or anything where the provenance of the answer is the product, that is the lane for a purpose-built tool.
A general model is built for open-ended reasoning. Framing an argument, drafting from a messy fact pattern, summarizing a deposition, stress-testing a position, translating dense analysis into a client-readable memo. These are judgment-shaped tasks where breadth and reasoning matter more than a warranted citation index. The general model is the sharp associate who reads fast and thinks broadly, with the caveat that it must be supervised.
The connectors are what make using both practical. Claude calling Westlaw or Harvey means the reasoning tool can pull grounded material when it needs it, instead of guessing. That is the architecture firms were improvising by hand a year ago, now offered as plumbing.
A routing rule a firm can actually adopt
The useful output of all this is not a vendor pick. It is a routing rule simple enough to put on one page.
Send to the vertical platform: citation-dependent legal research, structured review and diligence, and any workflow where the firm needs a warranted, auditable answer tied to authority.
Send to the general model: drafting, synthesis, argument development, client communication, and exploratory analysis where the value is reasoning, not a citation of record.
Let the connectors move work between them, so a drafting session in the general model can reach into the research platform for grounded authority rather than fabricating it.
Keep one rule above both: a human verifies every cited authority before it leaves the firm. The tool that surfaced the citation does not change who is accountable for it.
The line that does not move
It is tempting to read Claude's legal expansion as the moment a general model became safe for unsupervised legal work. It did not. The 2026 sanctions record, from federal appeals courts to the New York Appellate Division, fell on lawyers who trusted confident output and skipped verification. Better connectors raise the ceiling on what the tools can do together. They do not lift the verification duty off the lawyer.
That is why the routing rule ends where it does. Route by task to get speed. Verify every authority to keep the shield. A firm that does both gets the leverage of two strong tools without handing its judgment to either one.
Frequently Asked Questions
What did Anthropic change for law firms in 2026?
On May 12, 2026, Anthropic expanded Claude's legal capabilities with twelve new plugins and connectors linking Claude to Thomson Reuters CoCounsel and Westlaw, Harvey, Box, Everlaw, and DocuSign, so the general model can now reach directly into vertical legal tools.
Should a firm choose Harvey or Claude?
Increasingly the better question is which tool for which task. Use a vertical legal platform like Harvey for citation-bound research and structured workflows, and a general model like Claude for drafting, synthesis, and reasoning. Connectors let them hand work to each other, so "both, routed by task" is a viable answer.
Does connecting Claude to Westlaw or Harvey mean citations are now reliable?
No. A connector improves grounding but is not a verification step. A human must still confirm every cited authority, because 2026 sanctions for AI hallucinations fell on the lawyers who filed them, not on the software.
What is the simplest way to decide where work goes?
Route by task: citation-dependent and structured legal work to the vertical platform, open-ended reasoning and drafting to the general model, and keep a firm-wide rule that a person verifies every authority before filing.
Sponsored Training
Informational tool analysis for working professionals, not legal, medical, or financial advice. AI tools do not replace your professional judgment.