Oracle is turning corporate software over to AI agents

From generative AI to over 1,000 agents

Oracle has been fleshing out its AI strategy over the past 18 months. It initially focused on embedding generative AI to help users draft text, such as performance reviews or financial report summaries. From there, it set out to build 50 specific-purpose AI agents.

Today, that number has swelled to over 1,000 task-specific agents. To manage these agents, Oracle introduced AI Agent Studio within Fusion applications, allowing customers and system integrators, such as KPMG, Deloitte and PwC, to stitch agents together, modify them, and inject human-in-the-loop approvals for compliance.

More recently, in March 2026, Oracle unveiled a new generation of agentic applications spanning customer experience (CX), human capital management (HCM), finance and supply chain at the Oracle AI Summit in London.

Traditionally, ERP systems served as systems of record that logged operational history and enforced policy. But with agentic applications, instead of people having to work through the data and determine the next steps, users can just prescribe a set of objectives, goals and performance indicators, said Ng.

Using an accounts receivable department as an example, he said a manager could simply instruct AI agents to maintain a specific cash inflow level or target a specific days sales outstanding (DSO) metric. The agents will then work behind the scenes, analysing transactional activities and data to provide the manager with optimised strategies. Once a strategy is selected, the agents farm the required work out to human payment collectors, even generating call scripts based on a client’s payment history and ongoing disputes.

“In this case, the agents are working together to interpret and reason the data,” said Ng. “It helps to make decisions and execute them. It’s a much more autonomous way of thinking about how you drive your business.”

Governance and the future of per-seat licensing

Addressing concerns around governance and data security, Ng said Oracle’s AI agents work directly and interact with Fusion applications through application programming interfaces (APIs). “Whatever guardrails that already exist in the apps are observed at the AI layer,” he said. “Agents are just doing the job on behalf of a particular person with a security profile.”

Oracle is also bolstering its internal operations with these tools, using agentic AI to build applications, handle first-line customer support and replace complex rule engines by allowing developers to simply describe desired system behaviours in natural language.

But as AI agents take on more workloads, it raises questions about the future of traditional SaaS pricing, which has long been tied to the number of people or seats licensed to access the software. Currently, Oracle provides generous token allowances – around 30 million tokens across different large language models – for agentic applications without charging a premium for an AI version of its software.

Ng acknowledged that if agentic applications eventually allow a finance department of 100 people to operate with only five, Oracle’s pricing model will have to adapt. “Today, we are charging by seats because there’s still a need to have accountants running reports, so the seats are not going away yet,” he said.

“But down the line, let’s say in five years, if people start to reduce the number of seats, Oracle is not going to charge by the number of seats. But at this point, we’re not seeing a mass exodus of people from their contracts.”

Amid the potential of agentic AI, Ng pushed back against suggestions that core ERP systems will become commoditised.

“If you’re a $2bn company and you tell regulators that your general ledger is going to be all over the place, how do you go to the SEC and prove that you are financially compliant?” said Ng, referring to the US Securities and Exchange Commission. “That layer is not going away. It’s the experience and productivity layer, and how you interact with the data, that’s evolving.”