Oracle Adds Gemini Models to Its Fusion Marketing Agents

Oracle just gave its Fusion agents a second brain to pick from. On July 30, Oracle and Google Cloud announced that Gemini models are coming to Oracle AI Agent Studio for Fusion Applications, the same platform behind the prebuilt agents already running inside Fusion Marketing, Sales, and Service. If you've deployed Oracle CX for a client, this is the update that finally answers "which model is writing this thing?"

What Oracle and Google actually shipped

Two models are landing inside AI Agent Studio for Fusion Applications: Gemini 3.1 Flash-Lite and Gemini 3.5 Flash. They join the Cohere and Meta models Oracle already offers there. Flash-Lite is positioned as the cheap, fast option for high-volume, price-sensitive tasks. Flash is the heavier-reasoning model, and Oracle specifically calls out video and presentation generation as a use case for it, according to The Register's coverage. The same models also extend to embedded AI inside Fusion Applications and Oracle NetSuite, not just to agents a customer builds by hand in the Studio.

Oracle's applications chief Chris Leone framed it as giving customers "greater choice as they build and extend agents and agentic applications that reason through complex, real-world business challenges." Google Cloud's Satish Thomas put it in Google's own terms: the deal lets customers "use Gemini in the applications and agentic workflows they rely on to automate workflows, accelerate decisions, and drive outcomes." Read between the lines and the message is simple: Oracle isn't betting on one house model, it's betting on being the place enterprises plug in whichever model fits the job.

Where this lands inside Fusion Marketing

This matters most for Oracle's growing bench of prebuilt CX agents, several of them built for marketing teams specifically. The Program Planning Agent helps marketers plan, launch, and optimize cross-sell and up-sell programs. The Program Brief Agent keeps product, marketing, and sales aligned on campaign execution. Both shipped in February alongside the Program Orchestration, Buying Group, and Customer Insights agents, all prebuilt, all included in Fusion Applications at no extra cost, all running on Oracle Cloud Infrastructure.

Oracle's language about the Gemini rollout covers "embedded AI use cases" across Fusion Applications broadly, the same bucket these marketing agents live in. That's a strong signal a Gemini-backed campaign brief is coming, but Oracle hasn't published which prebuilt agent runs on which model, or whether admins get to choose per agent versus Oracle picking a default. The Register also flagged an open question worth tracking: whether the existing Cohere and Meta options stick around once Gemini is in the mix.

Why model choice matters for an FDE running Oracle CX

Up to now, if a client asked why their Program Brief Agent phrased something oddly, or ran a task slower or pricier than expected, the honest answer was "whichever model Oracle picked, and we can't see or swap it." Model choice changes the conversation from a black box to a lever you can actually pull. A high-volume, templated task like drafting first-pass campaign briefs is a Flash-Lite candidate on cost grounds. A task that involves reasoning across a messy briefing doc, or generating a presentation deck, points to Flash. That's the same price-performance tradeoff you already make when picking models on Vertex AI or Bedrock, just surfaced inside a low-code enterprise app instead of your own pipeline.

What to check before your customer flips the switch

• Ask your Oracle account team whether Gemini access requires a separate Google Cloud or Vertex AI agreement, or if it's bundled into the existing Fusion Applications and AI Agent Studio license.
• Confirm which specific marketing agents (Program Planning, Program Brief, or others) get Gemini as a selectable model versus which stay on Oracle's current defaults.
• If you manage custom agents in AI Agent Studio, check the model picker for the two new options before assuming they're live in your tenant. Rollouts for enterprise SaaS features like this typically stagger by region and pod.
• Watch for whether Cohere and Meta model options are deprecated or kept, since that affects any agent you've already tuned against a specific model's behavior.

For anyone deploying AI inside enterprise SaaS rather than a custom stack, this is the reminder that model choice is becoming a checkbox feature even in low-code platforms, and the FDE's job shifts from picking a model to knowing which one is actually running behind the button your client clicks.