Oracle Releases 25 AI Apps Built to Plan and Execute Business Tasks

Oracle has released 25 Fusion applications powered by Fusion Claw, a new runtime designed to carry out multistep business tasks. The company says the applications are available now and bring its Fusion Agentic Applications portfolio to 75. With permission, an application can go beyond proposing a plan and change business records, making customer approval settings a consequential part of the launch.

Work that does not end with a suggestion

The new applications cover accounting-exception investigations, workforce staffing plans, shipment consolidation and sales-territory planning. Natalia Rachelson, Oracle’s senior vice president of Fusion Applications and product management, told SiliconANGLE that earlier agents handled more focused goals, such as improving renewals. Fusion Claw is intended for work that requires optimization and repeated changes to a plan.

Consider Oracle’s staffing example. An agent could weigh qualifications, availability, labor rules and cost targets, then calculate scheduling options. If authorized, it could update the company’s records and reopen the plan when an employee calls in sick. That last step is the distinction Oracle is selling: the system is designed to revisit a task as conditions change, not merely produce a one-time answer.

Fusion Claw uses a language model for reasoning and planning, while conventional software handles calculations over large datasets and processes transactions. Oracle says this limits the use of expensive AI models. The initial release supports models from Google’s Gemini and OpenAI; customers cannot currently choose their own open, lower-tier models for Claw tasks.

25Claw-powered applications available now

Oracle says all 25 Fusion Claw-powered applications are available now.

75Total Fusion Agentic Applications

Oracle says the release brings its Fusion Agentic Applications portfolio to 75.

Where the permission sits

Customers can set objectives, permissions, risk thresholds and approval requirements for Claw tasks, Oracle says. They can have an employee review a proposed plan or let an application execute within delegated authority. Rachelson expects organizations to begin with more human review and authorize more work without individual approvals as they gain confidence.

Oracle says the language model runs in an isolated environment and cannot directly update Fusion business objects. After a task, the system generates an Outcome Receipt showing which policies it applied, the decisions it made and the transactions it performed. Rachelson says the receipt includes the full text of policies used, giving reviewers more context than a conventional audit log.

Claw can also work with third-party systems through application programming interfaces and agent-connection protocols, Oracle says. That reach makes the permission settings relevant beyond a single Fusion screen: a plan may draw on other systems before it becomes an authorized change. The receipt documents what happened afterward; customers still must decide which actions warrant approval beforehand.

The cost and the unanswered test

Fusion Claw carries no additional charge with Oracle’s agentic applications product, but that product is a separate purchase from a standard Fusion subscription. Customers also pay for the AI units their work consumes. Oracle’s approach of reserving model use for reasoning addresses one source of expense, but it does not establish what a completed staffing or accounting task will cost a customer.

Oracle has not provided measured cost savings or evidence of how the applications perform in production. Availability gives customers a way to try the workflows, but the central test remains open: whether the applications can repeatedly finish complex work well enough to justify their usage costs and the authority customers give them.