New runtime lets AI plan complex work while conventional software handles the repeatable computation, with controls and audit trails built in
Oracle has launched Fusion Claw, a governed agentic execution runtime for its Fusion Agentic Applications. It combines AI reasoning with deterministic enterprise computation so highly complex work can be completed economically and at scale.
Reasoning and execution, separated
Under Fusion Claw, AI models handle planning and decision-making while conventional software carries out high-volume, repeatable calculations and transactions. The runtime supports larger, longer-running work, continuous replanning and greater governed autonomy. Oracle says this also helps keep AI costs predictable, because the system doesn't call the model at every step of a process.
Governance built in
Every outcome operates within defined objectives, permissions, policies, approvals and risk limits, and audit-ready records capture the decisions and actions behind it. Oracle says the runtime enforces these controls through two built-in features, the Enterprise Operating Envelope and the Enterprise Trust Harness, so agents stay within their delegated authority. Customers can choose how much autonomy to grant, from requiring approvals up to governed full-auto execution.
25 new applications
Oracle is launching 25 Claw-powered applications across finance, HR, supply chain, sales and other areas. In finance, for example, the ledger application analyzes account variances and flags issues before close, while a collections application prioritizes receivables and carries out policy-permitted actions. Users can state a desired outcome in natural language, and agents work out and execute the steps.
Industry reaction
Oracle's CEO said the product moves enterprise customers from AI assistance to execution, with enterprise-grade governance. A Google Cloud executive said bringing Gemini models to Fusion Claw pairs frontier AI with the context and controls of Oracle's applications. An IDC analyst called it a logical next step in enterprise agentic AI. Another analyst said the approach should help with the unit economics of running AI agents, though analysts also noted some catches.
Customers and partners can create, extend and manage agents through Oracle AI Agent Studio for Fusion Applications.