Moat repriced
Thomson Reuters and the moat an agent could copy
Thomson Reuters · Legal & professional information · Canada
Thomson Reuters bought its way into legal AI early. Then a general-purpose agent shipped an open-source legal plugin, and investors repriced the whole category in a single session — while revenue kept growing.
−18%
Thomson Reuters in one session, 3 Feb 2026
−14%
RELX the same day — its worst since 1988
+9%
Thomson Reuters revenue growth, Q2 2026
A general-purpose agent repriced a vertical workflow moat.
The record
- On 3 February 2026, after Anthropic released legal-workflow plugins for its Claude Cowork agent, Thomson Reuters fell nearly 18% — on track for its biggest one-day loss on record — while RELX, owner of LexisNexis, fell 14% (its worst day since 1988) and Wolters Kluwer about 13%. (Reuters, Feb 2026)
- Thomson Reuters had moved early: it paid US$650 million in cash for Casetext, maker of the GPT-4-based legal assistant CoCounsel, completing the deal in August 2023. (Thomson Reuters, Aug 2023)
- The business has not cracked: second-quarter 2026 revenue rose 9% to US$1.95 billion and the full-year organic growth forecast was raised to about 8% — yet the shares were still down about 15% for the year. (Reuters via BNN Bloomberg, Aug 2026)
A moat built from content and workflow
For decades Thomson Reuters’ legal franchise rested on two layers. The first was proprietary content: Westlaw’s case law, headnotes and editorial annotation, accumulated over generations. The second was workflow — the research, drafting and review tools lawyers lived inside every day, sold as subscriptions that renewed almost automatically. Together they made one of the most dependable businesses in information services, valued by investors as something close to a utility for the legal profession.
Nor was the company asleep. In 2023 it paid US$650 million for Casetext and its GPT-4-based CoCounsel assistant, and it went on to fold generative AI across Westlaw and its tax and accounting products. By the second quarter of 2026, about 32% of its underlying contract value was tied to generative-AI products (Reuters, Aug 2026). On paper, this was an incumbent doing exactly what incumbents are told to do.
The day the workflow looked copyable
On 3 February 2026 Anthropic released plugins for Claude Cowork, its agentic workplace tool, including one for in-house legal work: contract review, NDA triage, compliance checks. The plugin was open source, and its outputs were meant to be checked by lawyers. It did not touch Westlaw’s archive. But it suggested that the workflow layer — the part customers pay to sit inside — could be assembled from a general-purpose model and configured by a legal team itself.
Investors did not wait for evidence. Thomson Reuters fell nearly 18% in a session, RELX 14% and Wolters Kluwer about 13%; a Schroders analyst described investors as “aggressively repricing these areas” (Reuters, Feb 2026). Two days later chief executive Steve Hasker argued that no new AI models “of any measure or importance” were entering legal research. The shares fell almost 6% more, leaving them about 22% below their pre-launch level (The Logic, Feb 2026).
A freeze in the multiple, not the ledger
What makes this a frozen case is repricing rather than collapse. Customers kept paying: revenue grew 9% in the second quarter of 2026 and the company raised its organic growth outlook (Reuters, Aug 2026). What fell was the premium investors paid for permanence — the assumption that a vertical software layer on top of proprietary content would stay scarce. Each new agent release reopened the question; another Anthropic launch in May briefly knocked the shares 5% lower (Reuters, May 2026).
The lesson for incumbents is uncomfortable. Thomson Reuters adapted early by buying an AI product, but the threat came from a horizontal platform that made the product category itself cheaper to build. Owning the content remains defensible; owning the interface is less so. The test now is whether Westlaw’s data and editorial judgment can be sold as the trusted layer that agents depend on, rather than as the place lawyers must go.
The lesson
Buying an AI product does not protect a moat if the moat was the workflow. When a horizontal agent can rebuild your interface, defend the layer it cannot reproduce — proprietary, verified content — and price it as the input agents depend on.
How we’re reading this
This is a repricing, not a demonstrated loss of customers: Thomson Reuters’ revenue and guidance rose through 2026, and the stock had already fallen about 22% in 2025 amid broader software-valuation worries before the plugin launched. The sell-off priced fear of future disruption, which may prove overdone; analysts disagree. We are not claiming Anthropic’s plugin has displaced Westlaw. Share moves differ by exchange and currency; figures are as reported.
Sources
- 01Anthropic’s new AI tools deepen selloff in data analytics and software stocks, investors say
Reuters (via Yahoo Finance) · Feb 3, 2026
- 02Thomson Reuters shares plunge as it defends AI strategy
The Logic · Feb 5, 2026
- 03Thomson Reuters reports higher second-quarter revenue, raises full-year forecast
BNN Bloomberg (Reuters) · Aug 5, 2026
- 04Thomson Reuters completes acquisition of Casetext, Inc.
Thomson Reuters · Aug 17, 2023
- 05
Compiled from public filings, court records, company statements and reputable press. Figures are attributed to their sources; allegations are labeled as such. Not legal or investment advice.