Google’s new legal AI exposes a bigger battle over the enterprise stack

Google Cloud unveiled Gemini Enterprise for Legal, a specialized AI solution aimed at automating various legal workflows such as contract review, regulatory monitoring, document drafting, and data discovery. This move indicates that the next phase of enterprise AI competition will focus on who can best specialize the stack rather than merely who can develop the most powerful foundation model.

The company also introduced Gemini Enterprise for Financial Services, another agentic AI solution tailored for financial professionals, marking the first offerings in Google's series of industry-specific, packaged solutions built atop the secure and fully governed Gemini Enterprise platform.

Gemini Enterprise for Legal and Thomson Reuters' Thomson model, despite their surface-level similarities in providing AI assistance for legal workflows, differ significantly in their underlying structures. Google's offering is an agentic system built upon its existing Gemini models, while Thomson Reuters took a unique approach by developing a proprietary model trained on its extensive proprietary professional content, including Westlaw, Practical Law, Checkpoint, and Reuters.

This proprietary model, Thomson, even outperformed leading models in benchmark evaluations. However, both companies are pursuing specialized AI in distinct ways.

On one hand, Google is leveraging its own AI stack, which includes global infrastructure, custom silicon, foundation models, and an AI-ready data platform. The company is enhancing these components with specialized skills, secure Model Context Protocol (MCP) integrations, third-party agents, legal tech partners, and robust control plane features for risk management, audit logging, and governance. This approach demonstrates that specialized AI can emerge by working above the foundation-model layer.

On the other hand, Thomson Reuters opted for a different strategy by creating its proprietary model, Thomson, which was trained on decades of content and expert evaluations. While this approach may provide a competitive edge, Thomson Reuters acknowledges that it will also utilize other models when they are better suited to specific tasks. This pick-and-choose strategy allows the company to maximize the strengths of various models for different applications.

In conclusion, both Google and Thomson Reuters are making significant strides in tailoring AI for professional domains, each employing different strategies. Google is building its specialization around its existing AI stack, while Thomson Reuters is focusing on a proprietary, trained model. While the model remains a crucial element in developing specialized AI, these ventures suggest that other, unique value-adds can also significantly differentiate specialized AI offerings.