Oracle showcases Fusion Claw, an autonomous agent platform that turns policy documents into executable JavaScript rules to manage hospital staffing and other workflows, aiming for deterministic governance-rule execution.

An outcome receipt records task details, authority, policy documents, models used, and changes for auditability, while raising concerns about readability, review, and potential misuse without proper governance of policy definitions and code.

A Dataiku/Harris Poll cited that 82% of CIOs say employees create agents faster than IT can oversee them, underscoring the need for strong change control and readable audit trails as organizations rely more on AI agents.

Kurapati advises starting small, proving ROI before scaling, noting that automated agent coordination represents a significant trust decision for a business.

Customers should treat compiled policy like production code—with ownership and testing—expire old memory when policies change, monitor approvers’ diligence, and require explicit written details on how full auto reacts to data changes before enabling it.

The system operates in two stages: a frontier model generates a probabilistic plan, then ordinary software executes it deterministically via a compiled policy that translates documents into code, using the policy as the controlling layer across tasks.

Two governance questions matter most: how cleanly policy translates into code and how easily actions can be reversed, guiding whether processes run fully automatically or under human oversight.

Oracle enforces a containment model: Fusion Claw can only call pre-registered agents and runs in a temporary per-task container, with no direct access to core systems to preserve security and traceability.

In a late-September demo, Fusion Claw achieved 99% staff coverage for a telemetry unit by simulating three staffing options and selecting the most efficient plan within authority—without explicit human approval for that specific plan.

The policy compiler problem is highlighted: even with deterministic code, ambiguous or changing policies can yield incorrect outcomes, so governance must enforce policy clarity, versioning, and revocation of outdated approaches.