From skills-based tool to legal AI assistant that works like a colleague
CoCounsel Legal AI is Thomson Reuters’ professional-grade legal AI assistant, rebuilt as an agentic system that completes complex, multi-step legal tasks through a single plain-language conversation, using proprietary legal content and models to deliver accurate, secure, verifiable work that consumer AI tools cannot reliably match. Thomson Reuters has opened early access to this next generation of CoCounsel Legal, calling it the most substantial rework since the product came with its purchase of Casetext in 2023. Starting this week, every existing CoCounsel Legal customer in the United States can flip a toggle to move between the current and new experience during the early-access period. More than one million professionals across 107 countries already use CoCounsel, which means this redesign is not an experiment at the margins—it is a bet on how AI for law firms should work in daily practice.
Why one-conversation, agentic AI matters more than another chatbot
The core idea of the redesign is blunt: prompts are for hobbyists; professionals need workflows. The first generation of CoCounsel Legal was organized around discrete skills and structured workflows, forcing lawyers to pick a skill, define steps, and craft long prompts. The new version behaves like a colleague. A lawyer describes the matter in plain language; the agent drafts a plan, works through the legal issues, pulls from the firm’s own precedents as well as Westlaw and Practical Law, drafts with citations, and adapts as new information surfaces. In other words, it turns a single conversation into an end-to-end workflow—planning, research, drafting, and revision—without the user managing the plumbing of “which skill” or “which tool” to use. For busy practitioners, that is not a UX tweak; it is the difference between AI that adds friction and AI that quietly removes it.
This shift from prompt-driven AI to fully agentic infrastructure means fewer steps and bigger outcomes. A smarter chat interface lets a single plain-language query handle complex multi-step legal work, so lawyers no longer need long-form prompts or specialist AI knowledge. For law firms weighing AI options, this matters more than flashy demos: a legal AI assistant that understands the task, plans work, and executes across tools is far more valuable than yet another consumer chatbot that responds only to the last message.
Fiduciary-grade AI: why Thomson Reuters is moving beyond consumer models
Thomson Reuters is explicit about the gap between consumer AI and what legal work demands: law needs fiduciary-grade AI, not generic black-box models. In the company’s words, CoCounsel Legal is built for work subject to regulatory oversight and duties of care, where accuracy, confidentiality, and accountability are paramount. Consumer chatbots can be useful for brainstorming, but they cannot guarantee citations, trace training data to vetted legal sources, or meet law firms’ security expectations. By contrast, CoCounsel’s agent has native access to Westlaw, Practical Law, and the expertise of 2,700 attorney-editors who help train and ground the system. According to Steve Hasker, this combination makes their system “the only one that reaches this fiduciary-grade standard” today. For firms signing engagement letters and malpractice policies, that claim is not marketing—it is table stakes.
Hasker argues that the onset of agentic AI—the move from single-turn generation to tools that autonomously plan and execute tasks—is as big a step forward as the first wave of generative AI itself. That shift, plus the performance of the new CoCounsel Legal AI in beta, is why the company accelerated the release: he describes it as the most successful beta he has seen for utility, accuracy, and reliability. Beta users went beyond polite approval; more than a few told the product team they “F#@%ing loved” the new CoCounsel Legal AI. In a world flooded with cautious pilots, that kind of response from practicing lawyers is a strong signal that the product is crossing from novelty into dependable tool.
Inside Thomson Reuters LLM strategy: model-agnostic, but proprietary where it counts
Underneath CoCounsel Legal AI sits a hybrid approach: model-agnostic architecture plus a growing proprietary Thomson Reuters LLM. The next-generation system was rebuilt on Anthropic’s Claude Agent SDK, selected after extensive evaluations for performance on legal tasks, and the engineering teams on both sides have built strong working chemistry. At the same time, Thomson Reuters has developed its own large language model for legal, imaginatively called Thomson, through a research lab acquisition and collaboration with academic partners. That model is now in advanced testing and is starting to outperform the latest general-purpose models on specific legal tasks. Hasker’s strategy is clear: keep using the best external models, but ensure the company can switch or blend them with its own Thomson Reuters LLM to fit legal workloads and risk profiles.
This is the opposite of a one-model bet. CoCounsel Legal AI is being built so parts of the agent can run on Thomson and others on Claude, with the freedom to adapt as technology and regulation evolve. The company is also engaging with major tech platforms to make CoCounsel reachable through APIs, connectors, and the Model Context Protocol, ensuring the legal AI assistant can live inside the systems law firms already use. That flexibility matters: law firms do not want to replatform their entire stack around one vendor’s consumer chatbot. They want AI for law firms that respects their data, plugs into existing tools, and keeps them in control of which model does what.
What this means for everyday legal work—and what comes next
For ordinary users, the impact is straightforward: CoCounsel Legal AI becomes one of the first places they go to get real work done. Starting the week of June 22, all current customers automatically unlock the next-gen experience, while new customers receive it from day one. During early access, users can switch back and forth between versions, but the direction of travel is clear: toward a single-conversation interface where a matter lives in an AI-powered workspace, with documents, precedents, and positions carried across sessions and colleagues. The reaction from beta participants is framed not as incremental improvement, but as a shift in expectations about what a legal AI assistant should do. For law firms still stuck in pilot purgatory with generic chat tools, CoCounsel’s adoption at scale signals that AI for law firms is moving from novelty to infrastructure.
Thomson Reuters plans full general availability of the next-generation CoCounsel Legal in August 2026 in the U.S., followed by rollouts in Canada, the U.K., and Australia. Workspaces—a dedicated environment for each matter that continues to evolve—is already live and will keep developing alongside other planned capabilities. The company’s Future of Professionals research warns of a growing gap between AI adoption and real value, and CoCounsel Legal AI is its answer to that problem: a system built around actual tasks, content, and accountability, not generic chat. The message to the market is pointed. Consumer AI will remain useful, but for fiduciary work, Thomson Reuters is betting that proprietary, verifiable legal AI will win—and it is moving quickly to make CoCounsel Legal the default agent for serious law-firm work.
