Oracle Fusion Claw the new grab for fully autonomous agentic AI in ERP

Architectural changes center on execution runtime

The apps run in execution runtime software that contains the OS and other code AI models need to execute tasks. 

"We realized that, in order to execute specialist work and much more complex, long-running work, we needed a different architectural paradigm," Rachelson said. 

The Fusion Claw was inspired by OpenClaw, an open-source autonomous AI agent, but differs in focusing on enterprise rather than consumer applications, according to Rachelson. 

In a press release, Oracle said the execution runtime supports "larger-scale and longer-running work, continuous replanning, deterministic execution and increased governed autonomy."

Among the new apps are the following:

• A ledger app that works across large volumes of accounting entries to reconcile information, investigate exceptions and apply the calculations and analysis needed to close the books for financial reporting.
• Workforce staffing AI that considers coverage and service requirements alongside constraints like skills, accreditations, labor rules and costs to build and adjust staffing schedules to meet new conditions.
• A shipping consolidation app that aims to cut shipping costs by modeling timing, capacity, cost and other constraints and carries out the best supported plan through governed execution. 

Natalia Rachelson

How Fusion Claw handles outcomes and governance

While the new AI agents can execute entire processes, people still have an essential role, according to Rachelson. 

"They have to know the business quite well to define proper business outcomes," she said. "They set the policies. They set the boundaries, they grant the authority to AI and they remain accountable for the business outcomes and overall execution." 

The Oracle release explained how it all works. After people define the objectives, authority and accountability, a frontier model -- one of the advanced AI models from major developers like OpenAI and Anthropic -- reasons, plans, learns and adapts. Fusion Claw then uses deterministic computation to execute the steps needed to reach the desired outcome. 

Oracle claims the approach enables more economical use of AI by applying it only to the reasoning stages and leaving the rest to deterministic computation that can be carried out less expensively and at higher volumes. 

Fusion Claw's governance is managed within an "enterprise operating envelope" that identifies key variables, including the organization's objectives, standard operating procedures, policies, permissions, risk thresholds and approval requirements, Oracle said. 

An "outcome trust harness" enforces the operating envelope in every outcome. Once an outcome is completed, a "receipt" serves as an auditable record of what happened, including the authority applied, the evidence used, the decisions made, the transactions executed, etc. 

Users can set a Fusion Claw app to fully execute autonomously or dial it back to provide quick assistance to human actors.

Full autonomy remains elusive

Two analysts briefed on and shown demos of Fusion Claw agreed that it represents a significant advance in agentic AI and addresses the concerns of CIOs and other C-suite leaders that the imprecise probabilistic reasoning of both agentic and generative AI doesn't suit the deterministic business processes that require exact calculations and precision often carried out in ERP.

"I think it is a milestone," said Mark Vigoroso, founder and CEO of The Enterprise Edge. "This is Oracle's attempt to have agentic AI do what I would call the thinking and the executing work."

Although agentic AI is nascent and previous applications were focused on tightly bounded, simple and linear steps, it nevertheless fulfilled the vision of transforming ERP from a system of record to a system of action, Vigoroso said. He said Fusion Claw takes things further, as it tackles more complex, multi-step, multi-threaded tasks and wrestles with tradeoffs, risks and prioritization the way a human would.

Fusion Claw tasks will run in isolation in a sort of container without open access to the internet or APIs, which is a good solution to the governance problem, Vigoroso said.

"They're going to basically run with very specifically defined role-based permissions. They're not going to be able to run wild," he said.

Keith Kirkpatrick, vice president of research at Futurum Group, called Fusion Claw an evolutionary step.

"What this announcement says to me is taking a very structured way to utilize AI to identify the best way to get work done, but doing it in a way that's clearly governed," he said.

However, while the Fusion Claw apps automate a larger part of workflows than previous apps, end-to-end automation is still a long way off, Kirkpatrick said.

"This is another stepping-stone in the agentic enterprise and automating all of these business processes," he said. "We're some ways away from that coming to fruition because there is so much at stake and so much risk involved."

Data must be accurate and verified, and coordinating multiple steps increases the chances of something going awry.

"It comes down to whether the organization understands what the limits need to be to operate safely," Kirkpatrick said.     

The 25 Fusion Claw AI apps run on Oracle's Fusion Cloud Applications, which include ERP, human capital management, supply chain and manufacturing, and customer experience. They join 50 previously released agentic AI apps.

Rachelson said they will be available in the fourth quarter and free with subscriptions to the Agentic Applications, which are priced separately from Fusion Cloud Applications. "We're trying to tie outcomes to consumption. The more AI delivers for you, there is a correlation to how many AI units you use."

David Essex is an industry editor who creates in-depth content on enterprise applications, emerging technology and market trends for several Informa TechTarget websites.