By Matt Bares, Signal Consulting
On January 30, 2026, Anthropic announced a legal plugin for Claude Co-work. Within 48 hours, Thomson Reuters lost 20% of its market cap. LexisNexis parent RELX dropped 14%. LegalZoom fell 15%. Roughly $300 billion in software value evaporated across the broader market in two days.
If you're a practicing attorney, you probably heard some version of this and moved on. Another AI headline. Another round of tech industry panic that doesn't touch the actual work of advising clients, drafting agreements, and showing up in court.
But the market wasn't reacting to a product announcement. It was repricing an entire category of software companies based on one question: if the companies that build the AI models can ship legal tools directly to end users, what are we paying the middlemen for?
That question applies to every legal AI platform asking for your firm's budget, your firm's data, and your firm's institutional knowledge. How you answer it over the next 12 to 18 months will shape your firm's competitive position for years.
To understand the shift, you need to understand two platforms that represent two fundamentally different visions of how legal AI should work. They aren't just competing on features. They're competing on who controls the work product your firm creates with AI.
Claude Co-work with the Legal Plugin is Anthropic's agentic tool. You give it access to a folder on your machine. It reads, edits, and creates files while executing multi-step tasks in a sandboxed environment. The legal plugin adds contract review, NDA triage, compliance tracking, risk flagging, and briefing capabilities. It connects to external tools like Slack, Box, Jira, and Microsoft 365 through MCP (Model Context Protocol), an open standard Anthropic created for connecting AI to external systems. Your playbooks are plain-text markdown files stored locally. Your firm defines its own standard positions and escalation triggers in a settings file you control. The AI comes to your data. Your data doesn't go to the AI.
The trade-off: there's no managed security layer, no firm-wide deployment tooling, and no support line. It's a research preview for Mac users on paid Claude plans. You're the admin, the implementer, and the quality control.
Harvey AI is the enterprise incumbent, valued at $8 billion after raising $160 million in December 2025. It offers a full platform with chat-based legal work, document storage, bulk analysis, and a Workflow Builder that has produced over 18,000 custom workflows across its customer base. It integrates with Microsoft 365, iManage, NetDocuments, and SharePoint. Harvey has a custom model built with OpenAI, layers in models from Anthropic and Google, and draws on more than 200 legal knowledge sources across 60-plus jurisdictions. You sign the contract, IT configures the integrations, and your attorneys start using it.
Harvey is turnkey. It's also a closed system. What you build inside Harvey lives inside Harvey.
The strategic question isn't which tool has better features. It's whether your AI tools let data and institutional knowledge flow freely across your technology stack, or whether they lock it inside a walled garden you can't leave without starting over.
Think about what happens when your firm spends two years building workflows, playbooks, and customizations inside Harvey. You've encoded how your firm reviews NDAs, how your M&A team runs due diligence, how your compliance practice flags regulatory risk. That institutional knowledge now lives inside someone else's infrastructure.
Can you export it? Can you plug a different AI tool into those same workflows? Can you take your playbooks with you if a better platform emerges next year?
For most firms, the honest answer is: you haven't asked. And the vendor isn't volunteering the information.
Harvey integrates with iManage, NetDocuments, and SharePoint. It pulls data from your document management system, your email, your file shares. But the workflows it helps you build, the playbooks it helps you codify, all of that lives in Harvey's ecosystem.
Claude Co-work takes the opposite approach. Your playbooks are text files on your machine. The skills files Anthropic released are open-source, published on GitHub, written in simple markdown that any capable AI model could read and execute. When Anthropic open-sourced MCP in late 2024, OpenAI, Google DeepMind, and Microsoft all adopted it within a year. Anthropic then donated MCP to the Linux Foundation's Agentic AI Foundation for vendor-neutral governance. The legal plugin skills follow the same path: portable, human-readable, not locked to any vendor.
The history of enterprise software is a history of lock-in that starts as convenience. Every SaaS platform begins by making your life easier. Then your workflows depend on it. Then your data lives in it. Then your switching costs are so high that you're no longer a customer. You're a captive.
There's a broader dynamic here that goes beyond legal tech. Across every industry, the software tools companies have relied on for two decades are facing a repricing. Not because they're going away. Salesforce isn't disappearing. Neither is iManage or NetDocuments.
But these tools are becoming infrastructure. A necessary but increasingly commoditized layer of the stack. The real value is migrating to whatever sits on top: the AI layer that works across all of your tools, pulls context from all of your data, and performs the knowledge work that used to require human judgment.
The question for law firms: who's going to own that layer for your practice?
If it's Harvey, they capture the value. If it's an open tool you control and customize, you capture the value.
This is why Anthropic's move rattled the market. Their legal plugin signals that the model providers are no longer content to be the plumbing underneath someone else's product. They're shipping domain-specific tools, bundling model plus workflow plus integration, and going directly to the customer. With open-source tools that anyone can customize and no one is locked into.
Harvey built its business assuming the model layer would stay neutral, that Anthropic and OpenAI would stay in their lane and let platforms like Harvey build the legal-specific wrappers on top. That assumption is being tested.
This doesn't mean Harvey is doomed. It has real advantages: enterprise security and compliance, firm-wide deployment, support teams, training programs, and deep relationships with AmLaw 100 firms. But its moat is shifting. It's less about AI intelligence and more about enterprise services. That's a very different value proposition than the one it started with.
If you're a solo or small firm (under 15 attorneys), start with Claude Co-work this weekend. The barrier to entry is a paid Claude subscription and a Mac. Download the legal plugin from GitHub. Point it at a folder of contracts. Run a compliance review. Triage some NDAs. You'll learn more about what AI can do for your practice in a few hours of hands-on work than any vendor demo will teach you.
If you're a mid-size firm (15 to 100 attorneys), you're in the hardest position. You need enterprise features like shared workflows, security governance, and compliance controls. But you can't afford to bet your firm's future on a platform that might get disrupted or might lock you in.
Consider a hybrid approach. Use Harvey for workflows that genuinely need firm-wide governance and enterprise security. Simultaneously invest in building internal capability with open tools. Identify the two or three people in your firm who are most curious about this technology and give them time to experiment. They're going to become your most strategically important employees.
And before you sign, ask your vendor three questions:
If you're a large firm (100-plus attorneys), you're likely already evaluating or deploying Harvey. The strategic question isn't whether to use it. It's how you avoid becoming dependent on it.
Build an internal AI strategy that treats Harvey as a tool within a broader architecture, not as the architecture itself. The firms that win long-term are building institutional AI capability they own: their own playbooks in portable formats, their own skills files, their own understanding of how to orchestrate agents across their entire technology stack.
Every technology wave in history has been deflationary. The cost of what software does for a law firm, per attorney, per matter, per unit of work, is going to come down. The era of paying $200-plus per lawyer per month for a platform that primarily wraps an API you could access for a fraction of the cost is going to face serious scrutiny. Open-source alternatives improve month over month. The model providers themselves are shipping increasingly capable tools at consumer price points.
On the other side of this equation: when an AI agent can complete 30 hours of document review in 20 minutes, the billable hour model doesn't get strained. It becomes indefensible. Corporate clients have been pushing for alternative fee arrangements for years. AI gives them the leverage to demand it. The firms that proactively shift to value-based pricing will capture the efficiency gains as profit. The firms that cling to hourly billing will watch their margins collapse as clients demand lower rates for work they know is being done by software.
Three hundred billion dollars in market value disappeared in two days. Not because legal technology companies are worthless, but because the market is repricing where future value gets captured.
The firms that build on open, portable foundations will have options. The firms that lock their institutional knowledge inside someone else's platform will have a vendor.
That's the decision in front of you.
Matt Bares is the founder of Signal Consulting, where he advises law firms on technology strategy and AI adoption.