Oracle launches Fusion Claw agentic runtime and 25 new applications

Fusion Claw combines large language models for reasoning and planning with conventional software for calculations and transaction processing. Oracle said the runtime is intended for longer-running work that requires extensive modeling, optimization and repeated changes to a plan, going beyond simpler tasks addressed by its earlier applications. Natalia Rachelson, Oracle's senior vice president of Fusion Applications and product management, said earlier applications could pursue focused goals such as improving renewals. 'We did not previously have solutions for advanced optimization,' she said.

In a staffing example, Oracle said a Fusion Claw agent might consider employee qualifications, availability, labor rules and cost targets, then propose scheduling options. It would evaluate the options, perform necessary calculations and, if authorized, make changes in the system of record. If an employee called in sick, it could autonomously reopen the plan.

Oracle said the Fusion Claw architecture limits use of expensive AI models. Claw uses a model when it needs to reason about a problem, while transactions and calculations over large datasets take place outside the model. The initial release supports frontier models from Google LLC's Gemini and OpenAI Group PBC. Rachelson said Oracle plans to support more models but did not provide specifics. Users currently cannot select their own open lower-tier models for Claw tasks.

The 25 applications include tools for investigating accounting exceptions, developing workforce staffing plans, consolidating shipments and planning sales territories. Oracle said Claw can also work with third-party systems through application programming interfaces, Model Control Protocol and the Agent2Agent protocol.

Giving software authority to change business records raises questions about oversight. Oracle said customers can set objectives, permissions, risk thresholds and approval requirements within what it calls an Enterprise Operating Envelope. They can choose a mode in which an employee reviews a plan or allows the application to execute within delegated authority. Claw runs in an isolated environment that prevents the language model from directly updating Fusion business objects, Rachelson said.

Once a task completes, an Outcome Receipt is generated showing the policies applied, decisions made and transactions performed. Rachelson said the receipt contains more context than a conventional audit log, including the full text of policies used during the task.

Oracle ultimately expects customers to authorize more work without individual approvals as they become comfortable with Fusion Claw. 'Honestly, any task can be in full auto mode,' Rachelson said, though she expects organizations to start with more human review and expand automation as they gain experience. The company did not provide measured cost savings or evidence of how the applications perform in production.

Oracle said all 25 Claw-powered applications are available now. Claw is included at no additional charge with the company's agentic applications product, which requires a separate purchase from a standard Fusion subscription. Customers pay for AI units consumed as work is completed.