Oracle Corp. is imbuing its Fusion applications with agentic artificial intelligence in a move aimed at shifting operational software from an assistive role into direct execution of business processes.

The company today unveiled Fusion Agentic Applications, a set of AI-powered applications embedded within its Oracle Fusion Cloud suite. They’re designed to reason, make decisions and take action across finance, human resources, supply chain and customer experience workflows.

The announcement reflects a broader transition in enterprise software that is seeing vendors attempt to evolve from systems of record into systems of execution. Oracle executives said the move builds on earlier investments in copilots and AI assistants but represents a more fundamental architectural change.

“We believe that AI is, especially for enterprise organizations, moving from advisers and copilots to being able to execute work,” said Chris Leone, executive vice president of applications development at Oracle.

Fusion Agentic Applications are built around coordinated groups of specialized AI agents that collaborate to achieve specific business objectives. Rather than responding to prompts or generating reports, they are designed to continuously improve outcomes by orchestrating tasks, evaluating tradeoffs and recommending or executing actions.

‘Team of specialists’

Leone compared the model to assembling a team of specialists to solve a problem. “Each agent has a specialty and the overall agentic application has a particular outcome that it’s trying to achieve,” he said.

The applications operate natively within Oracle’s transactional systems, giving them access to enterprise data, workflows, policies and approval hierarchies. Oracle said this tight integration allows agents to executive sophisticated workflows while maintaining governance and auditability.

Oracle is initially launching 22 agentic applications targeting use cases that span multiple business functions including workforce scheduling, supplier sourcing, cross-sell program management and cash collection.

Leone said the first set of use cases addresses what he described as high “cognitive load,” where work is often fragmented across systems or handled manually.

In the case of workforce management, “The agent can reason over all the information, understand context  and make recommendations as to specifically which absences and schedules to approve,” he said,

The agents are designed to operate with varying levels of autonomy. In a “human in the loop” mode, agents generate recommendations for approval by users. Organizations can also allow certain actions to be executed automatically, with the ability to increase autonomy over time.

“Customers can turn the dial up for more automation,” Leone said.

Build your own

A core component of the new features strategy is Oracle AI Agent Studio, which has been updated with a new Agentic Applications Builder. The tool allows customers to create their own agent-driven applications using natural language, assemble teams of agents and connect them to enterprise data and workflows.

The platform also includes capabilities for workflow orchestration, contextual memory, content intelligence and performance monitoring, including dashboards to measure the return on investment of AI-driven processes.

Oracle said the system supports integration with external platforms through application programming interfaces and the open-source Agent2Agent protocol developed by Google LLC. “We can send a query or question to another agent [in an external application] and get a response back,” Leone said.

Agents inherit role-based access controls and data permissions from existing application security frameworks in Fusion Applications. Oracle has also implemented auditability features to track changes to agent behavior and ensure accountability.

“If someone changes a system prompt  we can track that,” Leone said. In early testing, organizations have reported time savings of up to 40% to 50% in support scenarios, he said.

Oracle is adopting a hybrid pricing model for the new capabilities. Basic agents using built-in models are included with existing applications at no additional charge. More advanced capabilities powered by premium large language models will incur usage-based charges.

The company also plans to expand the ecosystem by enabling partners to build and distribute their own agentic applications, positioning the platform as a foundation for broader enterprise automation.

Leone said the shift toward agent-driven execution is likely to become standard across the industry. “We believe all systems of record will have to become systems of execution,” he said.

Image: SiliconANGLE/DALL-E