On Feb. 3, a lot of legal and data folks discovered they suddenly cared about stock tickers. Anthropic announced legal plug-ins for its agentic platform, Claude Cowork, and within 48 hours investors shaved tens of billions off the market caps of Thomson Reuters (Westlaw/CoCounsel), RELX (LexisNexis) and Wolters Kluwer.
Headlines framed it as AI killing legal software. Social channels called it the “Claude Crash” and the “SaaSpocalypse.” If you work in a Texas law firm or legal department, it felt like watching a tornado touch down a few miles over — not on your roof yet, but close enough to change your weekend plans.
Let’s separate the market drama from the operating reality. Yes, agentic AI changes how legal work gets done. No, it doesn’t vaporize the moat around verified content, citators and the editorial machinery that keeps the profession honest.
The center of gravity is shifting from search to verification. And in Texas, where procedural nuance and practice guides carry real weight, the winners will be the teams that pair new agent workflows with old-school authority and governance.
What Anthropic Released, in Plain English
Claude Cowork moved from chat bot to agent. The plug-ins let Claude plan and execute multistep legal tasks: think NDA triage against a playbook, clause comparison or first-pass issues tagging. Under the hood, the system talks to content and tools through an integration layer patterned on the Model Context Protocol.
Instead of waiting on custom integrations, firms can stand up MCP-style connectors to standard services, such as Google Drive, Microsoft 365, Slack, Box, Egnyte or Jira, and point agents at bounded content with specific skills and guardrails. The public repos and demos showcase legal-friendly tasks like playbook-driven review, summarizing inbound documents and organizing matter files.
Two things matter about this shift:
- It’s agentic, not just chatty. The value isn’t summarization; it’s orchestration. Claude plans steps, invokes tools, reads from approved sources and generates work product for human review.
- It lives where lawyers already work. Agents run in the environment you’re using — Claude’s app, or inside platforms that adopt MCP-like pipes — pulling the research and drafting layer into your workflow instead of sending you out to one more portal.
Why Investors Panicked
Public markets price narratives fast. The thesis was simple: When a general-purpose agent can stitch together steps across your content and SaaS tools, specialized “wrappers” around the same base models look replaceable. If you can vibe-code a contract triage workflow with a plain-English spec, why pay for a rigid app that does 80 percent of the same thing?
That logic moved stocks. Across two trading days, sector leaders took double-digit hits: roughly -17.8 percent at Thomson Reuters, -14.4 percent at RELX, -13 percent at Wolters Kluwer. Analysts called out margin pressure on seat-based software and the risk that “visibility premium” businesses lose pricing power when workflows migrate into general agent environments.
Here’s the counterweight. Law firm procurement moves at firm speed, not market speed. AI usage has to clear privacy, confidentiality, auditing and ethics reviews. Renewals are negotiated; migrations are planned.
And for all the talk of “wrappers,” the incumbents are more than just interfaces. They are data fortresses with editorial engines, citators and proprietary secondary sources that agents need in order to do legal work without getting you sanctioned.
The Durable Moat: Verification, Citators and Premium Secondary Sources
The core reality of agentic legal AI is simple: The more content an AI can produce, the more your job becomes verifying what it produced. Drafting speeds up; auditing becomes the bottleneck. That is why citators and trusted secondary sources got more, not less, important on the day Anthropic launched its plug-ins.
In Texas, the “crown jewels” aren’t abstractions. They are operational manuals you crack open all the time:
- O’Connor’s Texas Rules. Practical, deadline-driven guidance for civil trials in Texas. The difference between “almost” right and “filed on time” often lives here.
- Vernon’s Texas Codes Annotated. The statutory text plus Notes of Decisions — the curated, human-written links between a statute and how courts have actually treated it.
- Dorsaneo’s Texas Litigation Guide. Forms, strategy and procedural analysis that show up in court decisions and in your forms bank.
- Texas Practice Series, Texas Criminal Practice Guide, and Texas Torts and Remedies. Deep, frequently updated coverage for specialized and high-stakes work.
General models know the surface area of law. They do not know Texas practice until someone pipes in these sources and the citator layer that validates whether a case is still good law and how it’s been treated. That treatment analysis, such as KeyCite and Shepard’s, is not just table stakes. It is the safety system that keeps your filings grounded and your malpractice carrier happy.
There’s another moat, which is human editorial judgment at industrial scale. The West Key Number System and headnotes exist because attorney editors read, analyze and classify cases. Shepard’s treatment requires people to determine whether a case is affirmed, distinguished, criticized or overruled, and to layer that meaning across the citation graph. That editorial lattice is exceptionally well-suited to agent workflows because it gives the AI a concept map and a truth check, not just a blob of text.
A quick caveat: it’s easy to over-claim here. No citator is infallible, and anyone promising 100 percent is selling more than they can deliver. But there’s a reason the challengers publicly concede that their automated citators still trail the incumbents and that nine-out-of-ten accuracy is not good enough for risk-bearing work. In a verification economy, small error rates carry outsized consequences.
From Portals to Pipes: Protocols Reshape Vendor Strategy
If the last era was about destinations, meaning portals you logged into, this era is about pipes. The Model Context Protocol is functionally the USB-C of AI integrations, standardizing how agents connect to data sources and tools. The strategic fork for incumbents and startups is clear:
- Destination strategy: Keep users in your walled garden, bundle your AI, your content and your tools, and compete on end-to-end experience.
- Data-feed strategy: Expose your content and skills through MCP-style pipes so agents can call you where the user already is, then meter and monetize usage.
Both paths exist right now. MidPage, for example, leaned into the open-meadow approach by letting Claude call its case law via an MCP server, no website visit required. Incumbents have experimented with guided workflows (Thomson Reuters’ CoCounsel) and deeper integration layers (Lexis’s Protégé positioning) that bring their content and logic into the tools lawyers already use. The big strategic question for Westlaw and Lexis is timing and control: How far do they unbundle their data and tools into the agent ecosystem without commoditizing their interface?
Governance Pressure: Ethics, Security and Audit Trails
Texas lawyers don’t need hypotheticals to understand the risk side. Courts have sanctioned lawyers for fake citations. Texas courts and the bar have reminded everyone — repeatedly — that you, not your tool, are responsible for what you file.
Texas Ethics Opinion 705 put it in black and white:
- Independent verification is mandatory. You must check AI outputs for accuracy before you use them in client work. Blind reliance is an ethical violation.
- Technological competence applies. You need a working understanding of how your AI tools function, their limits and their data flows.
- Billing transparency matters. Don’t bill for time you didn’t spend; do address AI costs in engagement letters if you intend to pass them through.
On the security side, MCP is great because it lowers integration friction, but it is also risky for the same reason. Running agent tools against your repositories means you have to harden MCP servers, enforce least-privilege access, protect against prompt injection via tool calls and log every action. Your AI governance checklist needs to cover:
- Connector approval. Maintain an allowed list with security review for each MCP server and tool.
- Scoping. Define which repositories an agent can touch and under what conditions. “All drives” is not a scope.
- Logging and audit. Record prompts, tool calls, source citations and outputs. If the agent touched a document, you need a trail.
- Red-team testing. Treat prompt injection like a security vulnerability. Test for it. Patch for it.
- Human review gates. Require sign-off for filings, advice memos and anything that leaves the building under your name.
One more governance battlefield to watch is copyright and data licensing. The recent litigation over training use of editorial content (e.g., Thomson Reuters v. Ross Intelligence) reinforced that vendor-owned enhancements and secondary sources are not fair game for generic model training. If your agent needs the editorial layer, expect to pay for licensed access, and expect your vendors to enforce those boundaries.
How Incumbents Are Responding
The incumbents did not stand still. In the year leading up to the Claude Crash and immediately after, we saw a clear tilt toward workflow and integration:
- LexisNexis positioned the next generation of Protégé as a workflow layer embedded in where lawyers work, not just a research destination. The pitch is integration and guidance, not a chat box bolted on top of a database.
- Thomson Reuters pushed guided workflows around CoCounsel, connecting research artifacts to matter tasks and checklists. Again, the emphasis is on moving from a search engine to a “do this, then that” assistant with guardrails.
- Wolters Kluwer, always more diversified than it’s credited for, leaned into a broader portfolio: VitalLaw AI on the research side, along with enterprise legal management (ELM), tax and compliance platforms that don’t live or die on any one research interface. Diversification matters when the front-end category is noisy.
Each of these strategies acknowledges the shift from portals to pipes. Whether they ultimately succeed as destinations, as data feeds or as hybrids will come down to pricing, licensing and how well they instrument verification inside agentic workflows.
What This Means for Texas Firms and Legal Departments
Texas is not a footnote in this story. It is the center of gravity for how the profession balances speed and safety at scale. A few practical realities we’re already seeing on the ground:
- The hybrid is here. Administrative and business processes like intake routing, conflict pre-checks, calendaring, first-pass classification, are easy wins for agent workflows running in Claude or Microsoft 365. Substantive research and drafting stays anchored in Westlaw/Lexis and Texas-specific treatises, with agents pulling that authority in through pipes.
- Mid-market is nimble. Firms in the 50–200 lawyer range are moving faster than BigLaw because they can. With fewer committees and shorter procurement cycles, they’re vibe-coding internal tools, piloting reviewer bots and getting real productivity gains without pretending they can replace Dorsaneo or O’Connor’s.
- The associate role is shifting. The old “10 hours to research and draft” becomes “agent drafts in seconds; associate verifies against O’Connor’s and KeyCite/Shepard’s, adapts to judge’s preferences and explains risk.” Billing models will follow. So will training: Teach verification, synthesis and judgment first; teach mechanical research second.
- Contracts are the new battlefield. Your next renewal with a research vendor will read more like a data contract. Think API entitlements, MCP-connected usage, logs and audit rights. “Seats” won’t disappear overnight, but “pipes” will show up on the invoice.
- Shadow IT is both a risk and an opportunity. Vibe coding empowers practice teams to build useful automations fast. You need a lane for that energy, such as templates, approved toolkits and data access patterns, or you will end up with a dozen private agents crawling client folders with no oversight.
The Market’s Two Economies: Software vs. Data
The post-crash legal market is bifurcating:
- Commodity layer: Interface software and generic automations trend toward zero marginal value. If an agent can replicate a workflow in a weekend, you won’t pay a premium for a shrink-wrapped app that does the same thing.
- Premium layer: Verified data and editorial judgment trend toward premium pricing. The more content agents generate, the more valuable it is to have a ground truth that separates “sounds right” from “is right.”
That is why the “sell software, buy data” take spread so quickly after Anthropic’s announcement. It is a misreading to call the incumbents software companies. The asset that endures is the editorial graph tied to proprietary practice content. In a pipes world, that asset is incredibly portable and incredibly monetizable, if vendors price and package it well.
Security, Privacy and the MCP Elephant
A quick note for CISOs and privacy officers: MCP lowers the barrier to connecting an agent to your data. This is great, until somebody points the agent at “\all\share” and asks it to find “anything about merger holdbacks” across 30 years of files. The pattern to adopt:
- Small, scoped pilots. A single matter, a single team, a single repository.
- Clear data boundaries. Restrict read/write. Map to DLP policies. No production data in sandboxes.
- Vendor-by-vendor approval. If an MCP connector reaches out to an external tool, you vet the tool. Period.
- Prompt hygiene and policy. Treat agent prompts as data. They can leak client details as surely as emails. Log and protect them.
- Incident response for agents. If a tool misfires or a prompt injection triggers an unsafe action, know how you’ll stop it, roll it back and report it.
What About the Courts?
Sanctions for AI-generated hallucinations and local standing orders warning counsel to verify citations are already appearing. Expect more. Judges do not care if an agent wrote your draft. They care whether you signed your name to bad law. The verification economy will be enforced from the bench as much as from the bar.
Incumbent Countermoves and Pricing Reality
Expect three negotiations to define 2026–2027:
- API pricing and metering. If your “associate bot” runs a thousand KeyCite calls a day, what does that cost? Vendors will float per-call, per-matter and pooled firm-level pricing. Your usage logs will decide whether any of it is palatable.
- Content licensing scope. Can your agent store snippets? Cache headnotes? Use Dorsaneo excerpts in a drafting workflow? Where are the boundaries? The copyright lawsuits of the last few years mean vendors will be explicit, and firm counsel will need to be, too.
- Workflow bundling. CoCounsel and Protégé will arrive as bundles: content plus guided workflows plus verification. If you want just the pipes, you’ll need to ask for them — and to prove you can govern them.
Three Takeaways for Practitioners
- Agents change habits; verification still decides cases. Move fast on process automation. Move carefully on substantive law, and keep your secondary sources and citators front and center.
- Integrations are strategy. Value is moving from portals to pipes. Get your contracts, logs and IAM in order. Your next “research” renewal is an integration deal.
- Texas governance raises the stakes. Opinion 705, court sanctions and client expectations make sloppy automation expensive. Build review gates and audit trails into every agentic workflow.
Reporting and Voices You’ll Want in the Room
- Your firm’s research or KM lead: renewal terms, API rights and AI addenda.
- A Texas GC or legal ops head: where triage and compliance can safely go agent-first.
- A vendor product leader: connector roadmaps and how they’ll price per-call usage.
- Your security head: allowed connector lists, least-privilege defaults, logging standards.
- Ethics counsel tied to state bar guidance: where verification lines are bright and what can be delegated.
So, Did the Claude Crash Change Everything?
It changed what needed changing. We are moving from a world of “click here to search” to “tell me what you need, I’ll go do it, and you make sure it’s right.” The tools that do the doing are getting cheaper and more general. The sources that make “right” reliable are getting scarcer and more valuable. For Texas lawyers, that means two simultaneous investments: build the pipes that bring agents into your daily work and pay for the ground truth that keeps those agents from steering you into a ditch.
Don’t cancel your O’Connor’s and Dorsaneo. Do experiment with vibe-coded workflows for intake, classification and first drafts. Expect your research vendors to show up with new bundles and new meters. And when you see another spicy market headline, remember: Securities trade on stories, but law firms run on verified facts. In the verification economy, this is not a contradiction. It is the business model.
Greg Lambert serves as Chief Innovation Officer at Jackson Walker, where he leads the firm’s efforts to enhance service delivery through next-generation technologies such as generative AI, automation, and data-driven solutions.