The legal AI market has split into two distinct camps, and most law firms are picking the wrong one for the wrong reasons.

In one corner: Harvey AI. $200 million raised in March 2026. An $11 billion valuation. 70%+ of Am Law 10 firms already using it. Purpose-built for legal work, trained on legal documents, integrated into legal workflows.

In the other corner: Claude, Anthropic’s general-purpose AI, which now holds 73% enterprise market share across industries. Used by everyone from Fortune 500 legal departments to two-person boutique practices. Not built specifically for law, but doing serious legal work regardless.

The honest answer to which is better? It depends on your firm. And the answer matters, because the economics are wildly different.

What Harvey AI Actually Is

Harvey started as an AI assistant for lawyers and has evolved into a full platform for legal and professional services. Built on top of large language models (including, in part, Anthropic’s own Claude), Harvey has added a vertical layer of legal expertise, proprietary training on legal documents, and a platform architecture designed for enterprise law firms.

The core Harvey product has three components.

Harvey Assistant. A chat interface for legal work. Ask questions, draft documents, analyse contracts, get research.

Vault. A mass document analysis tool capable of processing up to 100,000 documents simultaneously. Built for large-scale M&A due diligence, portfolio analysis, and regulatory review.

Workflow Builder. A tool for building repeatable legal workflows. Harvey says its customers have created over 18,000 custom workflows, spanning everything from contract review processes to compliance checklists.

Harvey has also published genuinely interesting research on what they call “auto-research for legal agents.” The idea: agents edit their own harnesses (skills, hooks, sub-agents) to improve performance on legal tasks, without updating model weights. In their internal benchmark across 12 legal tasks, commercial lease review, complaint drafting, tax memos, due diligence, their optimised agents moved from an average of 40.8% task completion to 87.7%. Seven of twelve tasks finished above 90%. This is frontier work and it’s happening inside Harvey’s product today.

The catch: Harvey costs approximately £800 to £1,000 per lawyer per month. Minimum 20 seats. Minimum 12-month commitment. You’re looking at a minimum annual spend of around £230,000 before you’ve had a single meeting.

Claude is Anthropic’s general-purpose AI. Calling it “general purpose” undersells what it does in practice. Claude has a 200,000-token context window, which means it can read and analyse documents of up to roughly 150,000 words in a single session. For legal work, that’s transformative: entire contracts, bundles, or regulatory filings processed at once.

Claude is not a legal-specific platform. It doesn’t come with Vault, pre-built legal workflows, or a Workflow Builder. What it comes with is a genuinely powerful reasoning engine, excellent writing, and the ability to handle complex analytical work across any domain, including law.

In February 2026, Anthropic launched a Claude Legal Plugin as part of its new Cowork capability. It’s purpose-built for in-house legal workflows: clause-by-clause contract review against configured playbooks with colour-coded flags, NDA triage, compliance and vendor-agreement checks, contextual briefings, and templated responses. It doesn’t close the gap with Harvey’s platform entirely, particularly for high-volume external-counsel work, but it significantly narrows it for the everyday flow of most in-house teams.

Claude for Teams costs around £25 to £30 per user per month. For a firm of 20 lawyers, that’s roughly £6,000 per year, versus Harvey’s £230,000 minimum. The gap is approximately 38:1.

What Harvey Does That Claude Doesn’t

It’s worth being honest about where Harvey has genuine advantages.

Vault for mass document analysis. If you’re doing large-scale M&A due diligence, processing thousands of contracts simultaneously, Harvey’s Vault is purpose-built for this. Claude can process large documents, but it doesn’t have a document repository product at this scale.

Legal-specific training. Harvey has trained on legal documents, legal precedents, and legal workflows in a way that Claude, as a general model, hasn’t. This shows up in the specificity of Harvey’s outputs for complex legal tasks, particularly in areas like commercial lease review, where understanding what constitutes a market-standard clause versus a genuinely problematic provision matters enormously.

Enterprise integrations. Harvey integrates with the document management systems, case management platforms, and practice management software that large firms already use. Claude can do this too, but it requires configuration work. Harvey’s integrations come out of the box.

Agentic workflows for complex legal tasks. Harvey’s auto-research work, agents that self-optimise their harnesses on legal tasks, represents where their product is heading. This is genuinely differentiated from what Claude offers out of the box today.

For Am Law 100 firms doing high-volume M&A work, regulatory review, or large-scale portfolio analysis, Harvey’s advantages are material.

What Claude Does That Harvey Doesn’t

The case for Claude in a legal context is just as compelling, and for most law firms, it’s more compelling.

Accessibility. There is no minimum seat requirement. A sole practitioner can access Claude for less than £30 per month and do genuinely sophisticated legal work with it. Harvey has never been built for this market.

Flexibility. Claude is not a legal platform. It’s a reasoning engine. Which means you can use it for legal work on Monday, business development on Tuesday, and HR policy drafting on Wednesday. Harvey is built for legal and for legal only.

Privacy model. With Claude, documents stay on your machine. They are not uploaded to an external platform. For lawyers with strict confidentiality obligations, this matters.

Configuration over convention. A properly scoped Claude install configures Claude specifically for how your firm operates: your precedents loaded as context, your style guides ingested, your specific practice areas addressed. The output is a Claude that knows your firm, not a generic legal AI. This produces something Harvey can’t offer at its price point: a system that sounds like it was written by your team.

Cost headroom for growth. A firm that implements Claude at £6,000/year has £224,000 in headroom versus the Harvey minimum, to invest in other areas, or simply to protect margin.

The Real-World Signal: What Lean Practices Are Actually Doing

The most interesting data point I’ve encountered recently came from Zack Shapiro, founder of Raines LLP. He runs a two-person law firm serving 200+ startup clients. The firm is built on Claude from the ground up, workflows configured, precedents loaded, client communication structured through Claude Projects. His article on this reached 7.5 million views. Not because he’s an AI enthusiast. Because he built a firm that operates at a scale its headcount would not suggest is possible, and he documented it.

That’s not a fringe use case anymore. I’ve spoken to regional law firms across the UK who are approaching AI implementation the same way. They’re not comparing Harvey and Claude on spec sheets. They’re asking: how do we get operational, fast, without betting six figures on a platform decision we haven’t validated?

For those firms, Claude with a structured enablement process is the answer. Not because Harvey is bad, it isn’t, but because the risk-adjusted economics point clearly in one direction.

How to Choose: A Framework

Here is a simple framework for making the decision.

Choose Harvey if:

Choose Claude if:

Consider both if:

One more piece of context. The legal AI adoption data is striking: 42% of legal professionals are now using legal-specific AI tools, up from 21% 12 months ago. 94% of those using it report benefits. But firm-wide adoption is only at 34%. The gap between individual use and organisational implementation is the problem that matters, and it’s not a technology problem. It’s an enablement problem.

Lawyers are using AI. But firms haven’t built the shared context, the configured workflows, or the governance layer that makes AI genuinely useful at the organisational level. This is true regardless of whether the tool is Harvey or Claude.

The firms that close this gap, that move from individual experimentation to firm-wide operational capability, will serve their clients better, retain their best people, and protect their margin against the cost pressures already reshaping the market. Corporate legal teams cutting outside counsel spend is a real and accelerating trend. The firms that survive are the ones that adopt, not the ones that observe.

If You’re Thinking About Getting Your Firm Started with Claude

I run scoped engagements specifically for professional services firms. Discovery, group session, individual configuration by practice area (litigation, corporate, real estate, employment, whatever your mix is), shared Projects loaded with your firm’s documents and precedents, reusable workflows built per role, and a 30-day adoption plan.

Firms leave with Claude configured and operational, not a training programme and a handout.

If you’re a law firm exploring this, this is the fastest way from “we should probably do something about AI” to “we’re using it on live matters.”

Talk to me about getting started →

Alex Lockey is an operating partner working with professional services firms across legal, recruitment, and accountancy. He has been implementing Claude for leadership teams since 2025 and is pursuing Anthropic Certified Architect status.

Sources and further reading:

Enjoyed this? Join the newsletter.

One email a week. What I'm building, learning, and what's actually working. No fluff.