Oracle

Agents reach the customer's systems three documented ways. Inside the suite, they get secure access to the knowledge stores, tools and APIs of Fusion Applications. MCP support extends agent capabilities with third party data and tools through external systems. A2A agent cards let agents communicate and pass context to third party agents over standardized connectors. A credential store in AI Agent Studio manages API keys and authentication tokens, so agents reach external services without exposing them.

Sourceoracle.com/news/announcement/ai-world-oracle-expands-ai-agent-studio-for-fusion-applications-with-new-marketplace-llms-and-vast-partner-network-2025-10-15read 2026-09-17

Workflow orchestration coordinates multi step, multi agent execution with rules that control how work moves between steps, built in logic and human oversight. The Agentic Applications Builder composes Oracle, partner and external agents into workflows in a natural language environment without traditional development. Fusion Agentic Applications are teams of agents that plan, reason and execute across finance, HR, supply chain and customer experience rather than running a single task.

Sourceoracle.com/news/announcement/oracle-expands-ai-agent-studio-for-fusion-applications-with-agentic-applications-builder-2026-03-24read 2026-09-17

Agents are grounded in the customer's own material rather than the model alone. Oracle states the aim as turning siloed departmental knowledge into unified enterprise intelligence that grounds agents in current, accurate information. Agents get secure access to the knowledge stores of Fusion Applications along with unified enterprise data, workflows, policies and transactional context.

The March 2026 content intelligence release brings unstructured first and third party data together with transactional data so agents can act on it. Named agents answer from the customer's knowledge repository and fall back to the model only where the company has no content.

Sourceoracle.com/applications/fusion-airead 2026-09-17

Approval workflows and oversight controls are a named platform capability, and human oversight is one of the controls inside the new workflow orchestration rules. Agents execute within the customer's existing approval hierarchies, policies and permissions rather than around them.

The shipped agents show the gate in use. A salary change agent routes the transaction through the standard approval process with security validation. An approval agent presents a recommendation the approver can accept, have written up as a composed rationale, or override with their own reasoning to strengthen the audit record.

Sourceoracle.com/applications/fusion-airead 2026-09-17

Attestations and certifications appear on Oracle's own cloud compliance pages, which name ISO/IEC 27001 for information security management, ISO/IEC 27017 for cloud specific controls, ISO/IEC 20000-1, FedRAMP with a pointer to the authorizations on the FedRAMP Marketplace, HIPAA, Germany's C5, Australia's IRAP and Hosting Certification Framework, and France's HDS certification for health data.

On the agent layer, agents operate inside the suite's existing permissions and approval hierarchies, with governance and auditability named as platform components.

Sourceoracle.com/corporate/cloud-complianceread 2026-09-17

Monitoring and observability give real time visibility into agent behavior and performance in production, which is visibility into the agents themselves. Oracle names auditability and governance as built in alongside them. An agent ROI dashboard reports outcomes per agent across workflows, teams and business functions, so behavior is reported agent by agent rather than only for the whole estate. Retention terms and export to an external monitoring system are not documented.

Sourceoracle.com/news/announcement/oracle-expands-ai-agent-studio-for-fusion-applications-with-agentic-applications-builder-2026-03-24read 2026-09-17

Contextual memory is a named platform capability. Agents remember context across interactions, workflows and agent collaboration, only the memories relevant to a specific task are retrieved, and agents can share context with each other to coordinate. That documents persistence and scoping.

No surface is documented where a user or administrator can view, edit or delete what an agent has retained. The same capability also includes learning from user behavior, which is absorbed learning rather than state anyone can inspect.

Sourceoracle.com/news/announcement/oracle-expands-ai-agent-studio-for-fusion-applications-with-agentic-applications-builder-2026-03-24read 2026-09-17

Oracle publishes a named region estate of 41 cloud regions across 26 countries. Commercial regions sit beside an EU Sovereign Cloud whose regions (Frankfurt and Madrid) are operated entirely within the European Union and separated from the commercial estate to align with EU sovereignty and residency requirements. Government regions serve the United States, the United Kingdom and Australia, and Dedicated Region places a full region in the customer's own data center.

Region availability for Fusion Applications and AI Agent Studio is not listed separately.

Sourceoracle.com/cloud/public-cloud-regionsread 2026-09-17

Oracle's own prebuilt agents are listed across the Fusion suite, each with the work it does. Among them are agents that monitor contract health and margin risk, build renewal briefs, coordinate scheduling and absence cover, automate interview scheduling and flag stalled requisitions, audit change orders, and guide supplier onboarding against policy.

Preconfigured agent and agent team templates in AI Agent Studio are instantiated, configured and deployed, or used as the starting point for a customer's own. The AI Agent Marketplace adds Oracle validated partner templates seeded into the studio.

Sourceoracle.com/applications/fusion-airead 2026-09-17

Agents deploy into the Fusion workflow itself, and Oracle states that Fusion Agentic Applications do not have to wait for instructions. Several shipped agents are described as watching a data surface continuously. One monitors accounts and opportunities and highlights changes or risks, another monitors contract health and margin risk, and another monitors team learning needs to anticipate compliance gaps.

No schedule, webhook, event or monitoring configuration is documented, so the autonomous side is asserted agent by agent rather than specified.

Sourceoracle.com/applications/fusion-airead 2026-09-17

In its 15 October 2025 announcement, Oracle states that AI Agent Studio enables customers and partners to select the best performing large language model for their business needs, with support for OpenAI, Anthropic, Cohere, Google, Meta and xAI. The choice is the customer's across six named providers rather than vendor side routing. The March 2026 release adds multimodal capability, so agents process and generate images, audio and video as well as text.

Sourceoracle.com/news/announcement/ai-world-oracle-expands-ai-agent-studio-for-fusion-applications-with-new-marketplace-llms-and-vast-partner-network-2025-10-15read 2026-09-17

Oracle documents access to Fusion Applications agents from other Oracle applications and from external applications through the /invokeAsync API. Calls are authenticated through OCI Identity and Access Management with an OAuth 2.0 bearer token and a two way trust built from confidential applications on each side. A2A agent cards let external agents collaborate over standardized connectors. Fusion Agentic Applications are built with both no code and pro code capabilities, and Oracle publishes a public Fusion AI Studio GitHub repository for developers.

Sourcedocs.oracle.com/en/cloud/saas/fusion-ai/aiaas/access-agents-beyond-fusion.htmlread 2026-09-17

Testing and validation is a named platform capability. AI Agent Studio carries a built in evaluation framework to evaluate agent quality, performance and behavior, and a prompt playground supports testing and debugging of agent behavior before agents are scaled in production. Oracle describes the platform throughout as one for building, testing and deploying agents.

The mechanism underneath is not documented, since no test set, fixture, dataset or scoring output is described.

Sourceoracle.com/applications/fusion-airead 2026-09-17

No browser, desktop or computer control is documented, and Oracle ships no robotic process automation product of its own. Agents act through Fusion Applications data and functions, the suite's tools and APIs, MCP connections to third party systems, and A2A connectors to other agents. That is tool calling rather than operating an interface, so the question of what happens when a screen changes does not arise for this product.

Sourceoracle.com/applications/fusion-airead 2026-09-17