TL;DR — Key Takeaways
- Oracle Fusion Claw separates frontier model reasoning from deterministic execution for calculations, business rules and transaction processing.
- The runtime is available in 25 applications across finance, HR, supply chain and sales, bringing Oracle’s Fusion Agentic Applications portfolio to 75.
- Fusion Claw adds run-level permissions, approvals and audit controls as Oracle gives agents more authority to act inside enterprise applications.
Oracle has introduced Fusion Claw, a new execution runtime for its Fusion Agentic Applications that connects AI reasoning and deterministic computing for complex enterprise work.
Each Fusion Claw workflow begins with a frontier model, such as Gemini or an OpenAI model, that analyzes the task, develops a plan and adapts it as conditions change. Fusion Claw moves repeatable work like calculations, business rules and transaction processing out of the reasoning model and into deterministic execution. Oracle says the separation avoids using expensive reasoning models for every operation and can make complex agentic workflows more economical to run at scale.
Fusion Claw also adds a run-level framework for applying Oracle’s existing enterprise governance controls as agents take on more autonomous work. An “Enterprise Operating Envelope” defines policies, permissions, risk thresholds, approval requirements and escalation rules, while an “Outcome Trust Harness” applies those boundaries to each run across identity, data access, capabilities and actions. When the work finishes, an “Outcome Receipt” records the authority applied, evidence used, decisions made and transactions carried out.
Oracle is making Fusion Claw available in 25 applications across finance, HR, supply chain and sales. The additions bring its total number of Fusion Agentic Applications to 75. Among the new applications are those for reconciling ledger entries and investigating accounting exceptions, building workforce staffing plans under labor and cost constraints, modeling shipping consolidation options, and comparing sales territory configurations. Customers can choose different levels of automation, from AI assistance to fully automated execution within explicitly delegated authority.
Fusion Claw runs on Oracle Cloud Infrastructure and currently uses frontier models from Google and OpenAI, with plans to support more models over time. The company already supports a wider selection through AI Agent Studio for Fusion Applications, including models from OpenAI, Anthropic, Cohere, Google, Meta and xAI. Customers can use the studio to build or modify agents and multi-agent workflows that work with Fusion business data, transactions, APIs and workflows.
The new runtime takes inspiration from a core idea behind the open source AI assistant OpenClaw by giving an AI agent the tools and working environment it needs to carry out a task over multiple steps. Oracle Executive VP of Applications Development Chris Leone wrote before the launch that applying that approach to enterprise systems requires additional controls around authorization, transactions, approvals and auditability.
Fusion Claw brings persistent agentic execution into applications that already contain financial, workforce and operational data. The key adoption question is how much authority enterprises will be willing to delegate to agents acting inside those systems.
