In a recent discussion with Chris Leone, Executive Vice President of Applications Development at Oracle, Cloud Wars Founder Bob Evans took a deep dive into the company’s new Fusion Claw capabilities within Fusion Agentic Applications. Evans kicked off the conversation by asking Leone how Fusion Claw differs from other Claw technologies and what problems it’s designed to solve for Oracle customers.
“We set out to really change how enterprise applications behave and run,” said Leone. “We started with Fusion Agentic Applications, kind of merging AI and applications under one umbrella, and what Agentic Applications did and continues to do is reorganize how we deliver applications and solutions to our customers around an outcome.”
However, to solve higher-value, more complex enterprise problems — like those handled by financial, inventory, or supply chain planners — Oracle needed a different type of governed agentic execution capability for reconciliation, optimization, and simulation. That led to the creation of Fusion Claw, which Oracle designed to handle higher-value, longer-running tasks in a controlled, governed way.
Reasoning vs. Compute
Recognizing Oracle’s focus on separating frontier model reasoning from deterministic enterprise computation, Evans asked why that distinction is particularly important as customers attempt to automate complex work.
“I call it the three-body problem,” said Leone. “We see people, humans, owning the outcome, being accountable for the outcome, and setting that governance layer.
“Then agentic applications are defined around an outcome, and Fusion Claw does the hard work.”
Leone explained that Oracle separates the process into two parts: AI handles the reasoning, planning, and assembly of the necessary tools, while deterministic enterprise computation is completed within Fusion Agentic Applications.
Governance-First Approach
Leone described AI governance as “the most critical thing” for Oracle’s work with agentic AI. “Governing an agent harness or governing a Fusion Claw is an engineering problem, and so we’re engineering the whole governance layer into the product,” he said.
In practice, that governance layer includes an enterprise operating envelope that sits above the Agentic Applications, and defines standard operating procedures, policies, guardrails, and risk limits. This governance applies across Oracle’s Agentic Applications, and every Fusion Claw run is stamped with the current state of that operating envelope, creating an auditable record of how the work was governed. “If you have to go back and audit a particular result, that outcome receipt has the trust harness stamp in it, and you know exactly why that decision or set of decisions was made,” said Leone.
“You’ve got 25 of these full auto-capable [Claw-powered] Agentic Applications,” continued Evans. “Could you give us an example of a use case that best shows this practical shift from AI being an assistant [to] executing the whole end-to-end process?”
Using a supply-chain example, Leone explained how Fusion Claw can optimize complex manufacturing and supply-chain decisions, including maximizing truck utilization and adjusting production schedules. These capabilities are designed to help companies tackle optimization problems that were previously difficult to manage.
Leone described the approach Oracle is taking with Fusion Claw as a necessary evolution in enterprise AI. “You will have to separate reasoning from execution,” he said. “The system of record will always be and continue to be the authoritative source of the state of the business.
“It knows what the current status of orders is, what the current status of inventory is, but it’s this middle execution engine that you need to tackle, and that’s what we’ve done with this new solution.”
