What happened
Oracle announced the two capabilities at Oracle AI World in Las Vegas on October 26, 2026. Fusion Companion is described as an intelligent layer for Oracle Fusion Cloud Applications. Rather than asking an employee to find a particular application, transaction, workflow, approval, or AI agent, Oracle says the layer uses the employee’s business objective, enterprise context, data, and permissions to determine how work should be completed.
The announced scope is broad. According to Oracle, Fusion Companion can select and coordinate transactions, workflows, approvals, AI agents, and agentic applications. The company presents this as applicable both to simple tasks and to complex business processes. It says Companion can coordinate specialized agents, business processes, and Fusion Agentic Applications, including Fusion Claw-powered applications intended for deep analysis, planning, simulation, optimization, and long-running execution over large volumes of enterprise data.
Access is intended to extend beyond a single Oracle interface. Oracle explicitly names ChatGPT as a supported AI tool and says users will also be able to work through customer- or partner-built AI interfaces. Alongside those external entry points, Oracle is introducing Fusion Agentic Home, a personalized native AI experience within Fusion Applications. In Agentic Home, employees can define objectives, delegate work to AI agents and agentic applications, and track that work through completion.
Oracle’s core promise is governed execution. Fusion Companion is said to retain enterprise execution and accountability within Oracle Fusion Applications while using established identity, permissions, policies, and approval controls. The announcement also says it can use data and context from other enterprise systems, and coordinate authorized actions across Oracle Fusion, other cloud applications, enterprise content, data platforms, APIs, services, and third-party agents. For business exceptions, Agentic Home is described as identifying root causes, determining corrective actions, executing authorized resolutions, and escalating decisions that need human judgment.
The products sit on Oracle’s existing Fusion portfolio, which spans finance, HR, supply chain, manufacturing, sales, service, and marketing. Oracle also points to AI Agent Studio for Fusion Applications as the place where customers and partners can create, extend, and manage agents and agentic applications using enterprise data under the company’s stated security, governance, and performance posture. The announcement does not state a price, release date, regional availability, API specification, or a complete list of supported AI clients.
What the archive adds
This is a new fact in an already active Oracle Fusion story, not a restatement. On the same date, we covered Oracle’s Deep Trust AI Architecture and Fusion Claw. Those items described security and a governed execution runtime; Fusion Companion adds the employee-facing and external-client entry layer that Oracle says will invoke that governed system.
Trend signal
The announcement supports the pattern that agent-ready software exposes capabilities. Oracle is not merely positioning a conversational assistant inside a business suite: it says external and customer-built AI interfaces can reach Fusion work while Fusion determines authorized transactions, workflows, and approvals. It also supports agent security moves to runtime boundaries, because Oracle locates control in identity, permissions, policies, approvals, and the system where execution remains accountable.
Connections to other events
Fusion Companion is directly connected to the previously covered Fusion Claw governed execution runtime: Oracle explicitly includes Fusion Claw-powered applications among the components Companion may coordinate. It also meets the company’s Deep Trust AI Architecture announcement, which framed security, privacy, and business rules across Oracle’s stack. The new announcement supplies a user and client access surface for those related claims.
Why it matters
For teams building enterprise AI workflows, the important design question is increasingly not which chat client receives the request. It is where the request becomes a controlled business action. Oracle is proposing that employees may begin in a familiar AI tool, but that Fusion should retain the context and authority required to select transactions, run approvals, coordinate agents, and record accountable execution.
That is relevant to any team connecting AI interfaces to systems of record. A useful architecture needs to distinguish between an interface that interprets an objective and a runtime that has permission to act. Oracle’s announcement makes that distinction explicit in product positioning, even though it provides no implementation detail. Builders should also note the claimed cross-system scope: coordination across applications, content, data platforms, APIs, services, and third-party agents increases potential usefulness but makes authorization and review boundaries more consequential.
Limits and uncertainty
This is a product announcement, not technical documentation or an independent evaluation. Oracle does not disclose how Fusion Companion connects to ChatGPT or customer-built clients, which actions can be delegated in practice, which Fusion modules are supported first, or how third-party agents are authenticated and constrained. It provides no price, availability date, service-level commitments, usage limits, customer deployments, or measured productivity results. The stated ability to select, coordinate, and resolve work should therefore be read as Oracle’s announced product intent rather than evidence of performance on particular enterprise processes.
Skill to build now
Map one high-value workflow into explicit capability boundaries this week: identify the transaction to execute, data the agent may read, approvals required, exceptions that must escalate, and the person accountable for the result. Then separate the conversational entry point from the system that authorizes execution. That exercise is concrete whether a team uses Oracle Fusion or another system of record, and it exposes where an apparently helpful AI interface would otherwise receive too much authority.
Reader takeaway
Oracle is positioning Fusion Companion as a governed bridge between AI clients and enterprise business execution. The notable claim is not simply that employees can use ChatGPT with Fusion; it is that Fusion can remain the layer that supplies business context, chooses authorized work, applies approvals, and holds accountability. The missing test is whether those controls and integrations arrive with usable, documented interfaces.