Oracle: Simply give the AI the business goal

Contents

Oracle is enhancing its business applications with new-generation AI-powered features. Specifically planned are systems that not only provide answers or complete individual tasks but coordinate multiple specialized agents to autonomously achieve defined business goals. The new features are intended for use in Oracle Fusion Cloud Applications for ERP, HCM, Supply Chain, and Customer Experience (CX). To this end, the applications will run within the transactional core system and access business data, processes, approvals, policies, and process context.

According to Oracle, the offering is intended to differ from classic AI assistants: the applications are not only meant to provide recommendations but to independently initiate and execute actions within defined rules. At launch, Oracle names 22 agentic applications for workforce operations, procurement, sales, and finance. The AI agents are intended to optimize personnel deployment and payroll, manage suppliers, uncover cross-selling opportunities, or collect outstanding receivables.

Agent teams instead of individual assistants

In a press briefing, Natalia Rachelson, Senior Vice President, Cloud Applications Development at Oracle, described the approach: “The new agentic applications are teams of agents, where each agent has its own special role.” Users define a business goal, and the agents determine the path to achieve it. Examples include reducing outstanding receivables, identifying delinquent customers, or finding suppliers under specific conditions.

The applications use reasoning – the ability of AI models to draw multi-step conclusions – and context-aware functions to divide tasks, coordinate intermediate steps, and pass results on to users. Furthermore, Oracle intends to support open standards and external data sources, Rachelson said. These include the MCP (Model Context Protocol) and A2A (Agent2Agent, protocol for communication between AI agents) interoperability protocols, as well as the integration of structured and unstructured third-party sources. She cited weather data for logistics or LinkedIn data for personnel recruitment as examples.

In parallel, Oracle is expanding its AI Agent Studio. New features include an Agentic Applications Builder, workflow orchestration, Content Intelligence (automatic analysis of unstructured content such as documents or emails), context-aware storage functions (allowing agents to remember previous interactions), multimodal LLM capabilities, a prompt testing environment (Prompt Playground), and a dashboard for evaluating agent performance. This allows companies to adapt existing agents or build their own agentic applications.

Autonomy instead of user actions

Oracle describes its AI strategy as a shift from a “system of records” to a “system of outcomes”: applications are intended not only to manage data but to drive operational results. Whether this works reliably with today's AI models remains to be seen – the productive use of AI agents continues to pose challenges, for example, in governance. Oracle argues that the applications are enterprise-ready because they inherit all the security and governance mechanisms of the Fusion platform. According to Rachelson, the agents “inherit” role-based access controls, approval hierarchies, and policies of the respective user. Furthermore, all steps are traceable and auditable.

Pricing remains open. Oracle outlines a usage-based model based on so-called “Action Units.” Customers will initially receive a free quota; additional fees will apply for higher usage. Oracle did not disclose specific prices.

Conclusion

Oracle is positioning its Fusion suite as a platform for operational AI functions beyond classic assistance. The central claim: to deploy multiple specialized agents directly within the business system to achieve defined goals. How viable the approach is in everyday use will only become apparent during productive deployment, as will the question of how Oracle will assert itself against SAP, Microsoft, and Salesforce.

(emw)