From Dubai: Enterprise software has long been the domain of human decision-makers and slow, process-driven systems. This week in London, Oracle attempted to fundamentally shift that model by embedding autonomous AI agents directly into its cloud applications, claiming they can reason through complex business problems and execute decisions at speed and scale.
The pitch is straightforward: Fusion Agentic Applications operate inside Oracle's transactional systems with native access to company data, workflows, approval hierarchies, and governance frameworks. Rather than requiring human sign-off for every action, these agents handle routine work autonomously while escalating only exceptions and decisions where human judgment materially changes outcomes.
Twenty-two agentic applications launched this week. A Workforce Operations agent, for instance, automates scheduling approvals and flags payroll problems before they cascade. A Design-to-Source Workspace agent coordinates engineering, supplier, and sourcing decisions that traditionally live in separate departments. For finance teams, a Collectors Workspace agent chases unpaid invoices continuously, compressing the time companies must wait between billing and payment.
On paper, the efficiency gains are compelling. Early testers reported time savings of up to 40 to 50 per cent in support scenarios. Oracle positions this as a fundamental architectural shift from systems of record that merely store data toward systems that actively drive business outcomes.
A Caution Sign from Analysts
Yet seasoned observers are not yet convinced. Balaji Abbabatulla, a Gartner analyst covering Oracle, offered measured praise. "This sounds good, but be cautious. It doesn't necessarily look as glittery as it sounds," he cautioned. The critical challenges lie beneath the surface.
Data integration remains a hard problem. Organisations deploying these agents must stitch together information from multiple sources: legacy applications such as SharePoint repositories, non-Oracle databases, and systems from competitors like Databricks or Snowflake. While Oracle provides tools to perform this work, the process is manual. "There's no kind of autonomous way of synchronising these different data repositories in the background," Abbabatulla noted. For large enterprises already invested heavily in alternative data platforms, the transition overhead is massive.
More fundamentally, no vendor has satisfactorily answered the liability question. If an autonomous AI agent makes a cascading series of decisions at speed, causing costly errors or exposing the company to regulatory risk, who bears responsibility? Oracle's current answer is monitoring and audit tooling. Abbabatulla was unpersuaded: "I don't see a clear response from any vendor on the liability issue."
The problem is acute. Industry research suggests 80 to 85 per cent of enterprises lack clear liability frameworks for agentic AI failures. When an autonomous system operates with minimal human oversight, traditional product liability models break down. The harm becomes difficult to trace to a single design flaw or human oversight. Courts and regulators have not yet agreed on whether responsibility falls on the vendor who built the system, the organisation that deployed it, or some combination of both.
A Genuine Trade-Off
This is where genuine tension emerges between Oracle's commercial incentive and prudent enterprise governance. The vendor understandably wants customers to delegate authority to these systems because autonomy is where the value lives. Narrowly constrained agents cannot deliver the promised efficiency gains. But every expansion of autonomous authority increases the distance between human judgment and consequences.
Other analysts took a more bullish view. Mickey North Rizza, of IDC, called the shift "significant" and positioned Oracle as a market shaper in agentic enterprise software. The vendor's deep integration of agents into its applications, she argued, gives it structural advantages over point solutions or loosely coupled platforms. Enterprises facing board pressure to deploy AI agents at scale will find appeal in a tightly integrated suite.
For organisations weighing deployment, the practical calculus is complex. The efficiency gains are real and measurable. But the liability framework remains a patchwork of evolving standards, with different jurisdictions approaching accountability differently. Some jurisdictions favour strict liability regimes where deployers bear full responsibility. Others apply traditional negligence principles. Few have clear rules that apply specifically to autonomous agents operating at enterprise scale.
Oracle's announcement represents genuine progress toward autonomous enterprise systems. The engineering is solid, and early customer feedback is positive. But until organisations can clearly define who bears the cost when these agents fail, large-scale deployment will remain hampered by institutional caution. Vendors, customers, and regulators all have work to do.