Oracle Quietly Buried the Copilot: Its AI Now Lives Where Your Data Lives

Oracle is rebuilding the plumbing inside its Fusion ERP suite. Not with a chatbot on top — with agents that live in the same transactional system where invoices get coded, purchase orders get approved, and supply chains get re-routed when a supplier goes dark.

The company announced 22 Fusion Agentic Applications at Oracle AI World in London on March 24, spread across its ERP, supply chain management, human capital management, and customer experience clouds. The architecture is the story: these agents run inside Oracle Fusion Cloud Applications, with native access to the same data, policies, approval hierarchies, and governance frameworks that human workers use every day. That sounds obvious, but it is a meaningful departure from the copilot pattern — a chat window bolted onto an opaque system — that has dominated enterprise AI deployments so far.

"The execution, the typing of the invoices, the typing of the purchase order — that is what is going to be replaced in whole in AI," said Steve Miranda, executive vice president of applications development at Oracle, in comments to Reuters.

The distinction matters architecturally. A copilot queries data; an agent inside the transactional system can act on it. Oracle's agents can autonomously process invoices, generate purchase orders, flag exceptions for human review, and execute within the approval chains already configured in Fusion — without routing through a separate AI layer that lacks visibility into who needs to sign off and why. The company calls this being a "system of outcomes" layered on top of its existing "system of record."

The 22 applications span finance, HR, supply chain, and CX. In ERP, new agents include the Source-to-Settle Assurance Advisor, Record-to-Report Assurance Advisor, and Access Request Assistant, alongside expanded GenAI capabilities in enterprise performance management. SCM received the largest batch — more than a dozen new agents handling tasks like contract negotiation, cycle count analysis, and order configuration. HCM additions focus on workforce operations, scheduling, and applicant screening. CX agents now advise on pipeline, products, contracts, subscriptions, and compensation.

Oracle is also opening the stack. The Agentic Applications Builder inside Oracle AI Agent Studio lets organizations build and run automation using Oracle, partner, and external agents without traditional application development. The company did not specify which large language models power the reasoning layer, beyond saying they run on Oracle Cloud Infrastructure with "industry-leading LLMs."

The timing is loaded. Oracle's shares have fallen roughly 40 percent this year as investors bet that general-purpose AI agents could disrupt the market for specialized enterprise software — the kind that runs finance, HR, and supply chains at large companies. Oracle's bet is that its existing install base and deep integration into those workflows is exactly what makes it well-positioned for agentic execution, not vulnerable to it. Executives have made that case repeatedly since the beginning of the year.

"One of the persistent challenges with enterprise AI has been bolting intelligence onto existing workflows without deep integration into the transactional system," said Michael Fauscette, CEO of Arion Research, in Oracle's press release. "Oracle's approach with Fusion Agentic Applications is notable because the agents operate inside the application suite itself, with native access to data, policies, approval hierarchies, and the governance framework that enterprises require."

ISG's Mark Smith called it "a meaningful shift in enterprise software by moving beyond task automation to outcome-driven execution on the journey to an autonomous enterprise." IDC's Kevin Permenter added that agentic applications are most powerful when they "reduce noise that consumes people's time and elevate the decisions that need human judgment."

What Oracle is describing is not hypothetical. The Fusion Agentic Applications are shipping now, not on a roadmap. The 26A release cycle, which Oracle detailed on its Fusion Insider blog, delivers these agents across the Fusion Apps suite with specific capabilities — invoice handling enhancements, change order automation, cash basis accounting support, and embedded banking services via a partnership with Bank of America.

That's worth noting: the agents aren't confined to Oracle's own stack. The SCM agents can interact with data from third-party software connected to Fusion, which matters for enterprises that have accumulated years of heterogeneous integrations. How cleanly those agents handle heterogeneous data in practice is the kind of thing that will surface in customer deployments, not press releases.

Our read: Oracle is making a genuine architectural bet here, not repackaging a copilot. Whether that bet pays off depends on whether enterprise customers trust the agents enough to let them execute autonomously inside systems that touch real money and real supply chains. The install base is large enough that even partial adoption moves the needle — for Oracle's revenue trajectory, and for the broader question of when agentic enterprise software stops being a demo and starts being operational reality.