Oracle launches Fusion Claw, 75 agentic apps

Oracle's earlier Fusion applications pursued focused goals such as improving renewals. Fusion Claw is aimed at longer-running work that requires extensive modeling, optimization and repeated changes to a plan, as in a staffing example where an agent weighs employee qualifications, availability, labor rules and cost targets, then proposes scheduling options and, if authorized, makes changes in the system of record.

The architecture limits use of expensive AI models: Claw calls a model when it needs to reason, while transactions and calculations over large datasets happen outside the model. It supports frontier models from Google LLC's Gemini and OpenAI Group PBC, can work with third-party systems through APIs, Model Control Protocol and the Agent2Agent protocol, and runs in an isolated environment so the language model cannot directly update Fusion business objects. An 'Outcome Receipt' records policies applied, decisions made and transactions performed, including the full text of policies used.

How far customers go with automation is likely to hinge on comfort with that oversight setup. Oracle expects organizations to start with more human review and expand automation as they gain experience, and Natalia Rachelson said any task can be in full auto mode. The company did not provide measured cost savings or evidence of how the applications perform in production, so the practical test may be whether customers delegate authority once they see receipts and results.