Fusion Claw: Oracle’s New Runtime for Autonomous, Governed Enterprise Work
Oracle has introduced Fusion Claw, a new execution runtime designed to support more complex and autonomous work across Oracle Fusion Agentic Applications.
The technology is aimed at tasks that require extended reasoning, computation, simulation, modelling and continuous adjustment as conditions change. Oracle says 25 Claw-powered applications are available at launch, bringing the total number of Fusion Agentic Applications to 75.
The announcement represents another step in Oracle’s move towards agentic ERP. Earlier Fusion Agentic Applications focused on bringing AI agents into areas such as finance, human resources, supply chain and customer experience. Fusion Claw is designed to provide the underlying runtime for workloads that require agents to reason, plan and execute over longer periods.
Fusion Claw Separates AI Reasoning From Enterprise Processing
Oracle describes each execution as a Claw Outcome.
A run begins with an advanced AI model that can reason about a task, create a plan and adapt that plan as new information becomes available.
The execution itself is then passed to deterministic enterprise computing processes. These processes handle large volumes of transactions in a consistent and repeatable way.
The distinction is important for enterprise applications. AI models can provide judgement and reasoning where needed, while conventional computing handles high-volume transactional work where precision and consistency are essential.
Oracle says this approach could also improve the economics of agentic applications by limiting the use of AI reasoning to areas where it provides value, while using more efficient computation for large-scale processing.
Fusion Claw runs on Oracle Cloud Infrastructure and currently supports frontier models including Gemini and OpenAI models. Oracle says additional models will be supported over time.
New Applications Show How Claw Can Be Used
The initial applications demonstrate how Oracle expects the runtime to support complex enterprise processes.
Ledger can analyse large volumes of accounting entries, reconcile information and investigate exceptions ahead of financial close.
Shipping Consolidation evaluates different options across factors including timing, capacity, cost and service commitments. Once an appropriate plan has been identified and authorised, the system can move that plan into execution.
Workforce Staffing applies the approach to employee scheduling, while Account Territory Growth Plan uses it to support sales coverage decisions.
The common model is consistent across these examples. AI is responsible for analysing the situation and determining an appropriate course of action, while enterprise systems provide the transactional processing and business controls needed to execute the result.
Governance Is Defined Before Work Begins
Oracle has designed Fusion Claw around a governance structure that defines the boundaries of autonomous activity.
The first component is the Enterprise Operating Envelope.
This captures the rules within which an agent is allowed to operate. These can include organisational objectives, standard operating procedures, policies, permissions, risk thresholds, decision rights, approval requirements and escalation rules.
The second component is the Outcome Trust Harness.
It applies the operating envelope to each individual run and manages areas such as identity, capabilities, data access and permitted actions.
The third component is the Outcome Receipt.
Once a task has completed, the receipt records information about the run, including the authority under which it operated, the evidence it used, the decisions it made, transactions it executed and the resulting outcome.
For organisations considering greater levels of autonomous ERP activity, this audit trail could become an important part of governance and internal control processes.
Customers Control the Level of Automation
Oracle is not positioning every process for immediate full autonomy.
Customers can determine how much automation is appropriate for each workflow, ranging from quick assistance through to fully automated execution within explicitly delegated authority.
This creates a staged approach to agentic ERP.
An organisation could initially use an agent to analyse exceptions and recommend actions. Once the process has been tested and appropriate controls established, the same workflow could potentially move towards greater automation.
Oracle manages these capabilities through AI Agent Studio for Fusion Applications.
The Agentic Applications Builder allows organisations to create and manage agentic applications while providing capabilities for observability, return on investment measurement and safety controls.
What This Means for ERP Teams
Well-defined processes are the best candidates for automation. Fusion Claw relies on policies, decision rights, approval rules and operating procedures to establish how an agent can act. Organisations with clear and current process documentation will be better prepared to increase automation.
Audit teams should be involved early. Outcome Receipts provide a record of the decisions, evidence and actions associated with an autonomous run. Internal audit and finance teams should determine how this information fits into existing control and review processes before automation is expanded.
Start with assistance before moving to full autonomy. Organisations can begin with lower-risk use cases, such as analysing financial close exceptions or supporting operational planning. Results can then be measured before increasing the level of autonomy.
Define authority before enabling execution. The ability to allow agents to perform transactions creates a different risk profile from using AI for recommendations. ERP teams should establish clear boundaries around what an agent can change, which approvals remain mandatory and when a human must intervene.
Measure the value of each automation stage. Agentic ERP should not be judged only by how much work an agent can perform. Teams should measure factors such as processing time, exception rates, accuracy, control effectiveness and financial impact before moving to the next level of automation.
The Bigger Picture
Fusion Claw illustrates Oracle’s broader shift from AI assistants towards autonomous enterprise software.
The objective is no longer simply to help employees find information or generate recommendations. Agents are increasingly being positioned to analyse complex situations, make decisions within defined boundaries and execute work inside ERP systems.
The challenge for customers will be determining how much authority to delegate.
Oracle’s use of operating envelopes, trust controls and outcome records provides a framework for gradually increasing autonomy while maintaining visibility over what agents are allowed to do.
For ERP leaders, the technology therefore creates two parallel priorities. Organisations need to identify processes where autonomous execution can create measurable value, while also ensuring that policies, permissions, approvals and audit controls are mature enough to support it.
As Fusion Claw expands, the key measure will not simply be how intelligent the agents become. It will be whether enterprises can trust them to act within clearly defined boundaries and demonstrate exactly what happened when they did.