CoCounsel’s Reinvention: From Prompt Toy to Legal Colleague
The next generation of CoCounsel Legal is a purpose-built legal AI assistant that Thomson Reuters has rebuilt as an agentic system capable of handling complex legal tasks through a single natural-language conversation, grounded in authoritative content and designed for professional-grade accuracy, security, and inspectable reasoning.
This is not a minor update. Thomson Reuters has opened early access to what it calls the most substantial reworking of CoCounsel Legal since it acquired Casetext’s AI assistant in 2023, and every existing customer now sees a toggle to move between the legacy and new experience. Over one million professionals across 107 jurisdictions already use CoCounsel as their professional-grade AI tool, which means this redesign is not a lab experiment but a production change to how high‑stakes work gets done. This matters because the legal market has been flooded with general consumer AI; the new CoCounsel is a statement that serious firms need a different class of system. The takeaway: if your practice still relies on generic GPT-style tools, you are about to look outdated.
Inside the Agentic Architecture: One Conversation, Whole Matter
The defining promise of the CoCounsel Legal next generation is complex legal task automation in a single conversation. Instead of forcing lawyers to pick “skills” and chain prompts, a user describes a matter in plain language and the system drafts a plan, works through legal issues, pulls from firm precedents plus Westlaw and Practical Law, drafts with citations, and updates as new facts appear. The result is one iterable work product, not a pile of disconnected answers. For day‑to‑day legal AI assistant features, this is the difference between a clever search box and a junior colleague who owns the assignment from intake to draft.
Architecturally, Thomson Reuters is shifting from prompt-driven workflows to a fully agentic infrastructure that automatically chooses and sequences the right internal tools and workflows. That is why early beta participants could complete real litigation and transactional work in a single session and, in the words relayed by the product team, “F#@%ing loved” the experience. One quotable takeaway from the beta: “The new version of CoCounsel Legal is now the first tool I turn to get stuff done.” For firms, the implication is stark: productivity gains will accrue to those who adopt AI that plans and executes, not AI that waits for prompts.
Why Purpose-Built Legal AI Beats Consumer GPTs
Thomson Reuters’ leadership is blunt about why it accelerated this release: the arrival of agentic AI is as significant as the first generative AI wave, and the company already proved the underlying architecture through its Westlaw Advantage project. At the same time, AI has moved from experimentation to daily practice, and the gap between adoption and value is widening. In that context, next-generation CoCounsel is a bet that law firms need AI grounded in authoritative content and domain expertise, not generic models pointed at the open web.
The company says every citation in CoCounsel’s outputs is traceable to a linked authoritative source, reasoning is inspectable from the first step rather than after‑the‑fact checked, and confidential data is not fed into third‑party training pipelines. This directly answers a growing list of cautionary tales: even prominent firms have seen frontier consumer models hallucinate when they operate without curated legal content and embedded domain experts. The opinionated conclusion is hard to avoid: continuing to rely on consumer GPTs for serious legal drafting is now professionally reckless when purpose‑built legal AI is available.
TR’s Own LLM and the Strategy Behind Model Control
Under the hood, Thomson Reuters legal AI is deliberately model‑agnostic but increasingly self‑reliant. The next-generation CoCounsel currently uses Anthropic’s Claude Agent SDK, with TR’s teams calling Claude the best fit for their legal workloads after extensive evaluation. But the firm is not content to remain dependent: it is in advanced testing of its own large language model for legal, named Thomson, built from research roots that include a lab acquired around Cambridge and extended to Imperial College.
The plan is pragmatic rather than ideological. Some or all aspects of CoCounsel may run on the Thomson model while others continue to use Claude. That flexibility gives TR control over latency, safety, and legal reasoning depth, while preserving a strong partnership with Anthropic. For law firms, the strategic signal is clear: the future of AI for law firms is not about locking into a single general-purpose model but about orchestrating specialized models around trusted content and workflows. Vendors that cannot control their stack to that degree will struggle to match this level of assurance.
Workflow Impact and What Comes Next for Law Firms
This release is already in the hands of working lawyers. Starting the week of June 22, all existing customers automatically unlocked the new experience and can switch via a toggle, with full general availability planned for August 2026 and further rollouts to other markets afterward. Importantly, there is no migration project; firms can trial next-gen CoCounsel inside current environments. Integrations span Microsoft 365, HighQ, iManage, NetDocuments, SharePoint, Icertis, and several legal tech platforms, and the Anthropic MCP link means lawyers in Claude can reach CoCounsel without switching tools.
On the ground, feedback shows use from associates to partners across practice areas and across firms, in-house departments, and government roles. Specific legal AI assistant features on the roadmap include Brief Builder for agentic brief drafting with built‑in citation checks and issue spotting, plus a “firm and organizational intelligence” layer that lets teams encode their own playbooks into the system. In short, CoCounsel is being positioned as an agentic operating system for high‑stakes professional work, not a standalone chatbot. Law firms that treat this as optional innovation rather than an impending baseline risk watching their competitive advantage erode once again—this time, conversation by conversation.
