At the Oracle AI Summit in London this week, Steve Miranda, executive vice-president of Oracle applications development, outlined the company's vision for agentic artificial intelligence in enterprise software. Oracle recently unveiled Oracle Fusion Agentic Applications, a set of more than 20 AI-powered tools embedded across its cloud suites for enterprise resource planning (ERP), human capital management (HCM), supply chain management (SCM), and customer experience (CX).
Miranda explained that rather than attempting to automate entire business processes, Oracle is building agents that tackle very specific tasks. These agents run within tightly defined domains, constrained by business rules and data, and they surface only exceptions, trade-offs, and decisions where human judgement makes a material difference. The approach is designed to address common concerns among IT leaders about the risks of autonomous AI, such as incorrect outcomes or race conditions that could harm business operations.
Task-level autonomy with human oversight
Key to Oracle's strategy is the concept of task-level autonomy. Each agent is responsible for a narrowly scoped activity, such as analyzing cash collections risk, detecting real-time workforce scheduling gaps, or generating sourcing recommendations from design data. By keeping the scope narrow, Oracle can constrain the agent's behaviour more effectively than if it were operating across a broad, loosely defined process. 'Technically, this is a much easier problem to solve and much easier to have agent constraint,' Miranda said.
Moreover, Oracle incorporates a human-in-the-loop mechanism. Users can review the outcomes produced by the AI before they are executed, ensuring that the system remains a decision-support tool rather than an autonomous executor. This is particularly important for high-stakes functions like financial operations. Miranda noted that the impact of the agent is highly measurable: 'If you're using it to scan invoices, the AI agent can get it right or wrong. It's very measurable.'
Real-world impact: NHS Shared Business Services
Oracle provided a compelling use case through its partnership with NHS Shared Business Services (SBS). The organization uses Oracle Fusion Applications to standardize and automate financial operations across the National Health Service. According to Erika Bannerman, managing director of NHS SBS, the platform processes 7.1 million invoices, recovers £7.4 billion in debt, and handles up to £355 billion in NHS transactions per year. 'We've partnered with Oracle to provide a scalable, AI-powered platform that will help to transform how the NHS works so the focus stays on providing the highest standard of care for patients,' she said.
This example illustrates how agentic AI can deliver tangible efficiency gains in a mission-critical environment. By automating routine invoice scanning and flagging only exceptions for human review, the system reduces manual effort while maintaining accuracy and auditability. Oracle sees this as a model for how businesses can leverage AI to streamline back-office functions without sacrificing control.
Pricing and the future of enterprise software economics
The cost of AI is a pressing issue for both technology providers and customers. Oracle currently charges for its applications on a subscription basis and adds separate fees for AI functionality. Miranda acknowledged that this model may evolve over time. 'There will come a day where our pricing model for our base subscription changes from user based to some sort of transaction based pricing, or includes pricing regarding the size of company,' he said. However, he admitted he is 'not terribly concerned about that day' because if the agentic AI delivers productivity improvements, it is a win-win for both Oracle and its customers.
Miranda also addressed the broader impact on employment and business focus. 'Nobody is in business to run ERP,' he said. 'The more we can save them on the ERP side, the more they invest in what they do.' This philosophy aligns with Oracle's long-standing approach of making enterprise software less of a burden and more of an enabler for innovation.
Background and historical context
Oracle's move into agentic AI builds on decades of investment in cloud applications and artificial intelligence. The company has been embedding machine learning into its cloud ERP suite for years, but the launch of Fusion Agentic Applications marks a significant step forward in terms of autonomy. Previous AI features focused on predictive analytics and recommendations, whereas agents can now autonomously execute routine tasks within predefined boundaries.
The timing is important. Competitors such as SAP and Workday are also developing AI agents for their respective ecosystems. SAP has introduced Joule, an AI copilot that can assist with tasks across its cloud suite, while Workday has embedded machine learning into its HCM and finance modules. Oracle's differentiation lies in its emphasis on task-specificity and human oversight, which may appeal to risk-averse enterprises.
Challenges and governance considerations
Despite the promise, agentic AI in enterprise applications raises significant governance challenges. Ensuring that agents do not overreach or produce biased outcomes requires careful design of guardrails, audit trails, and escalation procedures. Oracle's approach of limiting agents to fine-grained tasks helps mitigate these risks, but it also means that broader process re-engineering still requires human intervention.
Industry analysts have noted that Oracle's strategy represents a pragmatic middle ground between fully autonomous AI systems and simple rule-based automation. By keeping a human in the loop, Oracle reduces the likelihood of catastrophic failures while still delivering meaningful productivity gains. The company's emphasis on measurability also helps customers quantify the return on investment, which is critical for justifying AI spending.
Another challenge is data integration. Oracle's agentic applications are designed to work within the Oracle ecosystem, but many enterprises operate heterogeneous environments with multiple vendors. Miranda acknowledged that Oracle sees its applications as systems of record within a larger ecosystem that may include both Oracle and non-Oracle AI agents. Interoperability will be key to widespread adoption.
Expanded view of Fusion Agentic Applications
The 20+ applications launched cover a wide range of business functions. In ERP, agents can automate cash collections risk analysis by analyzing payment histories and customer credit profiles. In HCM, workforce scheduling agents detect real-time gaps and suggest optimal shifts based on employee availability and skills. In SCM, AI-driven sourcing agents can analyze design data to recommend suppliers and materials. In CX, agents can assist with customer issue resolution by automatically pulling relevant information from multiple systems.
Each agent is built on Oracle's underlying AI infrastructure, which includes large language models fine-tuned for enterprise data. Unlike general-purpose chatbots, these agents are grounded in structured business data and follow explicit business rules. This makes them more predictable and easier to govern.
Miranda also highlighted the importance of continuous improvement. As the agents process more transactions and receive feedback from human reviewers, they learn to make better recommendations. Over time, the knowledge accumulated can be shared across similar tasks, potentially reducing the need for manual oversight in low-risk scenarios.
Broader implications for IT leaders
For CIOs and IT leaders, Oracle's approach offers a blueprint for how to deploy AI responsibly in enterprise settings. The emphasis on task-level autonomy, human oversight, and measurability addresses many of the concerns that have held back broader adoption of autonomous AI. It also provides a clear path to scaling AI from pilot projects to production use cases.
However, IT leaders will need to invest in data governance, change management, and training to fully realize the benefits. Employees accustomed to manual processes will need to learn how to interact with agents, interpret their outputs, and override them when necessary. Oracle is providing tools to help with this transition, but the onus ultimately falls on the customer to define appropriate use cases and boundaries.
Miranda concluded by emphasizing that the goal is not to replace human workers but to augment them. By offloading routine tasks to AI agents, organizations can free up employees to focus on higher-value activities such as strategic analysis, innovation, and customer relationship building. This vision aligns with the broader trend of AI-assisted work, where machines handle the repetitive and humans handle the complex.
As Oracle continues to roll out Fusion Agentic Applications across its customer base, the lessons learned will shape the future of enterprise software. The company's cautious yet ambitious approach positions it well to navigate the challenges of enterprise AI while delivering practical value today.
Source: ComputerWeekly.com News