Enterprise legal AI is not ChatGPT with a suit on
The next generation of CoCounsel Legal is an enterprise legal AI assistant redesigned to let professionals complete complex legal tasks in a single conversation, grounded in authoritative content, integrated with law firm systems, and engineered for accuracy, security, and verifiable answers rather than generic text predictions.
Thomson Reuters has opened early access to what it calls the next generation of CoCounsel Legal, the most substantial reworking of its flagship legal AI assistant since it bought Casetext in 2023. This is not a cosmetic upgrade; it is a statement about what serious enterprise legal software has to be. More than one million professionals across 107 countries and territories now use CoCounsel as their professional‑grade AI solution, and over 20,000 law firms and legal departments rely on it. When that many lawyers are involved, “good enough” answers from a generic chatbot do not cut it. Legal work carries regulatory risk, client exposure, and professional responsibility. Any legal AI assistant that cannot meet those standards belongs in the experiments folder, not in production.
From skills to a single conversation: what “agentic” really changes
The old CoCounsel Legal forced users to choose discrete skills and structured workflows one task at a time. That design mirrored consumer AI: you ask for a memo, you get a memo; you ask for a summary, you get a summary. Thomson Reuters has thrown that away. The next generation is built on a fully agentic infrastructure that asks a different question: what if completing any legal task required one conversation?
Now, a lawyer can describe a matter in plain language and CoCounsel Legal drafts a plan, works through the issues, draws from the user’s precedents plus Westlaw and Practical Law, drafts with citations, and adapts as new information appears. The system automatically determines which skills and workflows to engage and executes them in sequence. This is the opposite of prompt tinkering. For law firm AI tools, it means discovery prep, research, and drafting can run as a continuous project rather than a series of copy‑paste sessions between apps. For solo and boutique firms, it turns a legal AI assistant into that missing associate who can follow a matter from intake to draft.
Why enterprise legal software must be built for law, not the internet
Consumer tools like GPT are excellent at language and terrible at accountability. They pull from the public internet, which means results can be outdated, incomplete, or wrong for a jurisdiction. That might be acceptable when drafting a blog post; it is reckless when drafting a motion. CoCounsel Legal responds by grounding its work in Westlaw and Practical Law content curated by thousands of attorney‑editors, delivering answers tied to current law with citations lawyers can verify.
Enterprise legal AI also has to handle security and professional responsibility. CoCounsel Legal is not a repurposed consumer product; it is tested on real‑world legal tasks, operated to meet professional standards, and built to keep client data private and secure. Its integrations with Microsoft 365, HighQ, iManage, NetDocuments, SharePoint, Icertis, DeepJudge, Supio, and Smokeball turn it into infrastructure, not a toy. In other words, it behaves like enterprise legal software first and an AI demo second. Any law firm AI tools that do not meet that bar should stay far away from client files.
Why Thomson Reuters is building its own LLM – and why that matters
Under the hood, the next‑gen CoCounsel Legal runs on an agentic architecture built on the Claude Agent SDK, but Thomson Reuters is blunt about its long game: it does not want to rely only on third‑party models. The company is developing its own large language model for legal, called Thomson, already in advanced testing. CoCounsel is designed to be model‑agnostic so that some or all aspects can run on Thomson while others continue to use Claude, depending on which model performs best for a given task.
This is more than vendor hedge. Legal AI requires precision about sources, doctrines, and workflows; without access to authoritative content and in‑house domain expertise, models hallucinate. By training on its own content and using the same architecture and team behind its successful Westlaw Advantage system, Thomson Reuters can align the model with legal reasoning rather than generic prediction. That move also explains the timing. According to the company’s leadership, the onset of agentic AI tools is as significant as the first wave of generative AI, and their internal evaluations showed performance strong enough to accelerate release into limited access instead of waiting.
Beta love, real stakes, and what comes next
The early verdict from practitioners is not muted. During the beta, more than a few users said they “F#@%ing loved” the reimagined CoCounsel Legal, and one director at a major firm called it “one of the first tools I turn to when I want to get work done.” Feedback has focused on utility, accuracy, and reliability, with company leadership calling it the most successful beta they have seen. Crucially, this reaction is framed not as an incremental upgrade, but as a shift in expectations for what a legal AI assistant should do.
Starting the week of June 22, all current customers unlock the next‑gen experience, with new customers receiving it from day one. Early access will run with rapid, continuous improvements, followed by full general availability in late summer and a broader rollout planned for August in one market before expansion to others. The message is clear: AI has moved from experimentation to production, from pilots to daily practice. The firms that treat AI as a toy will be outpaced by those who treat enterprise legal software like CoCounsel Legal as infrastructure. In a world where consumer chatbots improvise, the future belongs to law‑grounded, agentic systems that can carry complex legal tasks end‑to‑end and back up every sentence with a citation.
