Oracle introduces Fusion Companion and Agentic Home for business tasks

Oracle announced Fusion Companion and Fusion Agentic Home on October 26, proposing a new way for employees to start work inside its business software. Instead of choosing an application and finding the right transaction, users describe the result they want. Oracle says the system can then coordinate the relevant processes and AI agents. Supported entry points include AI tools such as ChatGPT and Oracle’s own interface.

Fusion Companion supplies the connection between a request and enterprise operations. According to Oracle, it uses business information and the user’s permissions to select transactions, approvals and specialized agents. It can also coordinate authorized work involving other enterprise systems. The distinction matters for employees: asking for an outcome is only useful if the software can connect that request to the records and actions involved.

Agentic Home is the personalized interface built around that capability. Oracle describes users delegating tasks and following progress through completion, with decisions requiring human judgment escalated for attention. The announcement presents the two capabilities together but gives no general pricing or regional rollout timetable.

The execution technology underneath has an earlier history. Oracle introduced Fusion Claw on September 29, alongside 25 applications within a broader portfolio of 75. That announcement described combining model-based reasoning with conventional, deterministic computation: the model helps plan and adapt, while defined software performs work that needs repeatable calculations and execution. The October announcement therefore adds new entry points to a product direction already under development, rather than marking the first appearance of those 25 applications.

The September account also explains how delegated work is bounded. Organizations specify objectives, operating procedures, permissions, approval requirements and escalation rules. The runtime applies those limits to a particular assignment, and a completion record captures the evidence, decisions, transactions and result. Customers can choose different degrees of automation. Oracle’s description leaves the authority with the organization, even when the agent handles several stages of a process without repeated manual navigation.

A separate October 26 overview gives examples of the business work involved. The Ledger application is intended to investigate accounting exceptions and reconcile records, while Procurement Spend examines purchasing data for spending outside agreements and prepares corrective recommendations. Other applications address staffing, payroll comparisons and resource allocation. These are functions described by the supplier; the announcement provides no independent customer measurements showing how much time or money each application saves.

Oracle’s expansion also reaches the tools used to build agents. On October 21, it announced an Industry Applications edition of AI Agent Studio, with reusable agents, visual editing, testing and monitoring. That was a separate offering, accompanied by a plan for initial US availability within the following twelve months. It shows how Oracle is extending agent development beyond the Fusion interface while placing the different announcements on their own delivery schedules.

The industry announcement describes possible uses in construction and retail. A construction company could connect schedule risks with field issues and proposed mitigations; a retailer could extend work around product information and catalog management. Teams would configure the relevant data, roles, tools and policies, then test responses and actions before deployment. Those examples put the broader strategy in practical terms: the conversational request is the starting point, while business-specific connections and permissions determine what work can follow.