Oracle this week began rolling out Google’s Gemini models to its AI Agent Studio for Fusion Applications, extending the AI from development environments into the transactional workflows that run global businesses. The integration, first reported by IT Pro, gives organizations the ability to build AI agents that draft recommendations, analyze exceptions, and propose actions—while Fusion’s existing governance layer controls data access, approvals, and final execution.
What’s Changing: Gemini Lands Inside the Business Apps You Run Every Day
Oracle stated that customers and partners will now be able to access two specific Gemini models through AI Agent Studio for Fusion Applications: Gemini 3.1 Flash Lite and Gemini 3.5 Flash. The former is engineered for high-volume or latency-sensitive tasks; the latter tackles more complex reasoning and multimodal jobs, including video and presentation creation. These models join an existing roster of AI providers within the studio, giving development teams a wider palette for automating finance, HR, supply chain, and customer experience workflows.
Beyond the studio environment, Oracle plans to weave Gemini into embedded AI features across its Fusion suite—Enterprise Resource Planning (ERP), Human Capital Management (HCM), Supply Chain & Manufacturing (SCM), and Customer Experience (CX)—as well as NetSuite. In each case, Oracle will select the model that delivers the best price-performance for the specific scenario, a model-selection approach that elevates AI from a one-size-fits-all bolt-on to a tuned engine inside core business operations.
The key architectural detail: AI Agent Studio is already tightly integrated with Fusion’s knowledge stores, APIs, and business controls. That means a developer or partner can build an agent that, for example, flags a collections anomaly or a procurement exception, reasons over it using Gemini, and then routes its proposed action through the same approval chains, role-based access controls, and compliance checks that govern ordinary Fusion transactions. The agent does the thinking; Fusion remains the control plane for the doing.
Why This Matters for Windows-Centric IT Organizations
If your shop runs Windows-based infrastructure alongside Oracle—think Active Directory forests, Microsoft Entra ID, Power BI reports touching Fusion data, SQL Server integrations, or custom .NET apps that call Oracle APIs—the arrival of Gemini-powered agents demands your attention. These agents won’t live in a sandbox. They will, by design, read data, invoke APIs, and initiate processes across connected systems. Every endpoint that an agent can reach is a surface you must govern.
For identity and access teams: A Gemini-driven procurement agent might need to check a supplier’s status in an on-premises SQL Server database connected via a Windows gateway. That path must be audited and controlled—who can build an agent that traverses it, what data it can touch, and how its actions are logged. If you use Entra ID for federation, synchronize user attributes from Active Directory, or enforce conditional access policies, you’ll need to map how agent identities propagate across these trust boundaries.
For data integration specialists: Fusion agents could pull additional context from a Power BI dataset or a legacy Windows-based line-of-business application. You’ll have to classify which datasets are safe for AI consumption, set data retention policies for agent interactions, and ensure that prompting flows don’t inadvertently expose sensitive information to an external model—even one hosted within Oracle’s governed environment.
For app owners and business process experts: The promise is tangible. A financial close agent could analyze thousands of journal entries, identify outliers, prepare adjustment drafts, and route them to the right approver—shaving days off a cycle. But that only works if the agent’s training, prompts, and guardrails are configured to your FP&A team’s standards. Model choice isn’t just an IT decision anymore; business stakeholders must weigh in on which tasks warrant a fast, cheap model like Flash Lite versus a deeper reasoning model like Flash.
How We Arrived at the Application-Layer AI Shift
Oracle’s move is the logical next step in a multi-year journey. The company first opened Gemini access through Oracle Cloud Infrastructure (OCI) Enterprise AI, where it functioned as an infrastructure service. That was a developer playground. Now, by embedding models directly into the application layer, Oracle is responding to enterprises that want AI to work inside the guardrails they’ve already built—not as an external chat app that employees paste spreadsheets into.
The broader industry context is clear: Microsoft has poured Copilot into Dynamics 365 and the Power Platform, SAP is weaving Joule into its ecosystem, and Salesforce touts Einstein. Oracle’s countermove leverages its deep tie to Google Cloud, whose Gemini models have proven competitive on both performance and cost. By making model selection a per-use-case decision, Oracle subscribes to the growing consensus that enterprise AI is not a single model battle but a portfolio game.
Notably, the NetSuite component brings this capability to small- and mid-market firms that may lack the deep AI governance teams of Fortune 500 companies. Here, the promise—and the risk—is that guided wizards and pre-built templates will shield users from the underlying complexity. Oracle has not yet published a detailed feature-by-feature rollout or a firm timeline, which means early adopters will need to test carefully.
Your Action Plan: Steps to Take Now
-
Map your Oracle–Windows integration points. Inventory every connection where Fusion or NetSuite exchanges data with Windows-based systems: identity stores (AD, Entra ID), databases (SQL Server), reporting tools (Power BI), custom middleware, and file shares. Document these touchpoints and classify them by risk level if an AI agent were to read or write through them.
-
Strengthen identity governance for non-human actors. Determine how your identity provider will authenticate and authorize agent actions. If agents use service principals or managed identities, ensure those are scoped to the minimum necessary permissions and that their activity is logged to your SIEM. Review Federation trusts between Entra ID and Oracle Cloud to confirm that agent traffic does not bypass conditional access policies.
-
Establish an AI use-case review board. Before Gemini agents are turned loose, convene business and IT stakeholders to define approved use cases, data-handling rules, and acceptable model choices. For example, an agent processing credit limit adjustments might require a full audit trail and a human-in-the-loop checkpoint, while a simple shipment-tracking notification agent can run with lighter supervision.
-
Engage Oracle on the roadmap and sandbox early. Reach out to your Oracle account team to understand the timeline for Gemini availability in your Fusion and NetSuite instances. Request a sandbox environment where you can trial the AI Agent Studio with Gemini without touching production data. Use this to benchmark response times, evaluate output quality, and test how easily an agent can be constrained to approved actions.
-
Train your Windows sysadmins and Fusion administrators together. The silo between infrastructure and application teams breaks down here. A Fusion admin configuring an AI agent should understand the Windows authentication it relies on; a Windows admin must grasp the business impact of a bad agent decision. Joint tabletop exercises or lunch-and-learn sessions can prevent misconfigurations that lead to data leaks or process disruptions.
What Comes Next
Oracle’s announcement leaves open questions that will shape the real-world rollout. How will the embedded Gemini features surface in Fusion’s UI—as a side panel, a recommended action, or an autonomous background process? What telemetry and logging will Oracle expose to let admins monitor agent decisions, token usage, and model drift? And how quickly will the NetSuite integration deliver the same level of governance that Fusion promises?
The bigger takeaway: enterprise AI is shifting from a developer tool to an application-administration responsibility. Model choice is joining patch management, identity configuration, and data classification as a core IT competency. For the Windows professional, that means the next performance review may ask not just how many servers you manage, but how well you governed the AI agents that now help run your company.