Google's Finance and Legal Agents Shift the Build vs. Buy Math for Enterprises

Google Cloud launched industry editions of its Gemini Enterprise platform for financial services and legal work on August 25, according to Google's announcements for financial services and legal. Both are in preview. The move puts Google alongside Anthropic and OpenAI in the business of selling packaged, industry-specific agents, not just models and toolkits. For CIOs and chief AI officers in regulated industries, that changes the build-versus-buy question, and shifts the hard work toward data access, permissions and evaluation.

What did Google launch for finance and legal?

The financial services edition, initially aimed at capital markets and corporate banking, centers on a Google-managed Financial Research agent. Google says the agent provides confidence scores, explicit methodologies, auditable data snapshots and source citations, and can be called from other agents through Agent-to-Agent (A2A) APIs and connected to enterprise data through Model Context Protocol (MCP) integrations. It comes with more than 50 prebuilt skills for tasks such as credit risk assessment and portfolio monitoring, and 13 connectors to licensed data sources including FactSet, LSEG, Moody's, MSCI, PitchBook, S&P Global and SEC EDGAR. Partner agents from Dun & Bradstreet, FlowX, Obin and S&P Global are also available.

Google said CME Group is using the financial edition, and that Deutsche Bank, a design partner for the research agent, plans to deploy it starting in its corporate bank.

The legal edition includes skills for brief drafting, citation verification, contract lifecycle management and regulatory horizon scanning, plus connectors to systems such as iManage, NetDocuments, RelativityOne, Docusign, Harvey, Legora and Thomson Reuters, Google said. Launch customers include legal teams at Cleary, Freshfields, Weil and Williams & Connolly. Google said connectors inherit firms' existing permissions and ethical walls, which matters for law firms that must keep client matters separate. Pricing has not been disclosed for either edition, TechRadar noted.

Is Google catching up or changing the game?

Analysts mostly framed the launch as catch-up. Anthropic released Claude for Financial Services more than a year earlier and offers finance and legal plugins for Claude Cowork, while OpenAI offers industry plugins, including finance, on its Codex platform, AI Business noted. TechRadar added that Microsoft's Legal agent for Word reached general availability in mid-August.

David Nicholson of Futurum Group told AI Business that "Google is playing catch-up in a lot of ways," while noting that Google is leaning on openness to compete. Google Cloud CEO Thomas Kurian pitched the product as not tied to a single model or ecosystem, PYMNTS reported. Nicholson argued that the neutrality claim is weakened by Google's own stake in the model race, though companies already on Google Cloud may trust its guardrails more than a standalone lab's.

Sid Nag of Tekonyx offered a different reading in the same AI Business piece: Google's real asset is its connections to licensed financial data, not Gemini itself. That is the more useful frame for enterprise buyers. When every provider has a capable model, the vertical editions compete on which systems they connect to, how well they respect existing permissions, and how much evidence they show for each answer.

What does the vertical agent wave mean for build versus buy?

Many banks and law firms built their own research and drafting assistants on general models. Packaged editions now cover much of that ground, which raises a fair question for internal AI teams: what is still worth building?

The answer depends on where differentiation actually sits. A research agent that reads FactSet and SEC filings is available to every competitor on the same terms. What vendors cannot package is a firm's own data: its client history, internal risk definitions, deal records and the business rules that sit in people's heads. At a AI Connect webinar on August 26, Actian senior engineer Betty Wang warned that "AI will act on the data with total confidence whether it deserves it or not," arguing that definitions, lineage and freshness have to be in place first. Our coverage of that session on business context as the foundation for AI-ready governance lays out the metadata agents need.

That suggests a split. Buy the packaged agent for commodity research and drafting. Build, or at least own, the context layer that connects any agent to proprietary data, along with the evaluation sets that prove it gives the right answer for your business. The advice from our data executive's guide to AI use case selection still holds: tie each deployment to a measurable outcome before choosing a vendor.

What should regulated enterprises check before adopting?

Packaged does not mean production-ready. Before moving a preview into real workflows, leaders in finance and legal should test several points:

• Permission inheritance. Confirm that connectors enforce existing entitlements in practice, including edge cases such as matter-level walls or restricted research. Snowflake's Sam Mittal made a related point in a AI Connect webinar on agentic-ready data: human analysts catch numbers that look wrong, but "Agents don't do that. They take the data at the face value and act."
• Evidence and audit. Google emphasizes citations, confidence scores and data snapshots. Check whether those artifacts meet your model risk and record-keeping requirements, not just whether they exist.
• Data licensing. Connectors to licensed sources usually depend on your existing subscriptions. Clarify who pays and what usage terms apply when an agent queries data at machine scale.
• Portability. A2A and MCP support should make it easier to swap components. Test that claim early by connecting at least one agent or data source from outside Google's ecosystem.
• Evaluation. Build a test set from your own past work, such as credit memos or contract reviews, and measure accuracy before and after rollout. For a bank, a KYC and AML analytics workflow is a practical place to start, because the right answers are already documented.

The bottom line for enterprise leaders

Vertical agents are becoming a standard product category, and competition among three large providers should help buyers. But the models and packaged skills will look similar across vendors. The firms that gain most will be those whose own data, definitions and controls are ready for an agent to use.

Frequently asked questions

What is Gemini Enterprise for Financial Services?

A preview edition of Google's Gemini Enterprise platform for capital markets and corporate banking, with a Google-managed Financial Research agent, more than 50 finance skills and 13 connectors to licensed data sources.

Who is using Google's new finance and legal agents?

Google says CME Group uses the financial edition and Deutsche Bank was the design partner for its research agent; legal launch customers include Cleary, Freshfields, Weil and Williams & Connolly.

Should enterprises build or buy vertical AI agents?

Buying packaged agents makes sense for commodity research and drafting, while firms should own the context layer and evaluation sets that connect agents to proprietary data.

Filed under: Agentic AI, AI Use Cases & Applications, AI Strategy & ROI