Oracle Puts Gemini Models Inside Fusion AI Agents

Gemini Turns Up Inside the ERP

Oracle announced on July 30, 2026 that it is making Google's Gemini models available inside Oracle Fusion Applications AI Agent Studio, the platform its enterprise customers use to build, connect, and run AI agents against their own business processes. Oracle also plans to use Gemini for embedded AI features across Fusion Applications and NetSuite.

• Gemini models become selectable inside Fusion Applications AI Agent Studio, the environment where Oracle customers create and operate their own agents
• Thousands of Oracle enterprise applications customers gain access without a separate procurement or integration project
• Oracle plans to use Gemini for embedded AI use cases in Fusion Applications and Oracle NetSuite, not only in customer-built agents
• An expansion of an existing Oracle-Google partnership, following earlier multicloud and database collaboration between the two companies

Why Model Choice Inside the Application Layer Matters

There is a persistent gap between where models live and where work happens. A finance team's actual process sits inside an ERP — approvals, reconciliations, procurement records, close checklists — and the data that makes an agent useful is locked in those objects with their own permissions, audit trails, and workflow states.

Getting a frontier model to that data usually means an integration project: an API layer, an identity mapping, a security review, and someone owning it forever. Making the model a dropdown inside the tool that already holds the process removes all of that. The agent runs where the records are, under the permission model that already governs them.

That is the same structural point behind Cognizant's rollout of Claude to 40,000 enterprise staff — the adoption bottleneck in enterprise AI is rarely model capability, and almost always the distance between the model and the system of record.

What Can Customers Build With This?

Agents that touch the parts of a business nobody demos. Matching invoices against purchase orders and flagging the exceptions worth a human look. Drafting the narrative sections of a quarterly close from the ledger data. Summarizing a supplier's history before a contract renewal. Triaging expense reports against policy.

None of that is glamorous, and all of it is the kind of work that consumes enormous amounts of skilled time precisely because it requires context that lives in several places at once. An agent with native access to the application's own objects can assemble that context without a human doing the assembly first — which is the actual value, more than the language generation on top.

Google Cloud vice president Satish Thomas framed the goal as simplifying the use of Gemini in enterprise applications and workflows to accelerate decision-making. Our AI coverage has watched Gemini push steadily into working software this year, including Gemini building entire decks inside Slides.

Does Multi-Vendor Model Access Help Customers?

Yes, and the reason is unglamorous: it prevents a decision from becoming permanent.

Oracle already offers OpenAI models to enterprise customers through its universal credits programme. Adding Gemini means the agent a customer builds today is not welded to one vendor's roadmap, pricing, or availability. Given how quickly model capability and price have moved through 2026 — with several frontier tiers repricing within weeks of each other — the option to switch the model behind an agent without rebuilding the agent has real value.

There is a discipline cost. Multiple model providers behind one platform means prompt behaviour, token accounting, and safety characteristics vary depending on selection, and teams have to test against the model they actually deploy. That is a manageable problem and a better one to have than a rewrite.

The Shape of the Partnership

This extends a relationship that already spans multicloud infrastructure and database work between Oracle and Google Cloud. What is different about this step is that it reaches the application layer, where Oracle's customers spend their day, rather than the infrastructure layer they mostly do not think about.

For enterprise buyers, the practical read is that model access is quietly becoming a platform feature rather than an integration project — and the platforms that already hold the business data are the ones best positioned to offer it.

Sources: Oracle newsroom — July 30, 2026; PR Newswire — July 30, 2026; ITPro — July 2026.