Mike Sicilia is one of Oracle’s two chief executive officers, with a background in applications and industry software. His side of Oracle’s AI strategy is to embed agents in transactional systems such as finance, human resources, supply chain, and healthcare. The advantage is access to business context and permissions; the risk is allowing generated output or automated actions to enter systems of record without adequate review.

This profile was checked against public sources on September 14, 2026. Oracle product capabilities and customer outcomes are treated as company claims unless independently verified.

The co-CEO structure

Oracle promoted Mike Sicilia and Clay Magouyrk to CEO roles in September 2025, while Safra Catz became executive vice chair. Oracle’s investor announcement describes Sicilia’s responsibility for applications and industry operations and Magouyrk’s infrastructure background.

The 2026 Form 10-K confirms that both are chief executive officers and directors. It says Sicilia joined Oracle in 2009, previously led Industries and Global Business Units, and earlier worked at Primavera Systems. The SEC filing is the authoritative source for the current titles and audited company results.

Oracle does not publicly split authority into a complete decision map. It is reasonable to analyze Sicilia through applications and industry software because that is the documented experience and appointment rationale. It would not be sound to infer private reporting lines, compensation motives, or internal disagreements.

Applications give agents transaction context

Oracle Fusion applications contain financial records, employee workflows, procurement rules, supply-chain data, and approval hierarchies. That makes them a natural place to run narrowly scoped agents. An agent can potentially retrieve context and prepare or execute a transaction without a separate integration layer.

Oracle introduced Fusion Agentic Applications in March 2026, describing coordinated agents that can operate within application permissions and workflows. The product announcement explains intended capabilities and governance. It is a vendor source and should not be read as proof that every listed task is generally available or delivers the claimed outcome in every deployment.

Embedding AI also raises the consequence of error. A poor answer in a chat window is inconvenient. An incorrect journal entry, purchase action, benefit decision, or clinical-documentation suggestion can create financial, employment, or patient risk. High-impact actions need scoped identities, deterministic validation, approval thresholds, complete traces, and a tested reversal process.

Oracle Health raises the evidence standard

Sicilia’s industry remit includes Oracle Health, built around the Cerner acquisition. Clinical and billing workflows contain protected information and affect care and reimbursement. An AI feature in this setting should be evaluated for accuracy by specialty, editing burden, downtime behavior, consent and notice, security, and the distinction between documentation support and clinical decision-making.

The HHS summary of the HIPAA Security Rule explains the administrative, physical, and technical safeguards required for electronic protected health information. Compliance depends on the regulated entity’s deployment and contracts; a vendor’s security feature list does not by itself establish compliance.

Financial scale does not isolate AI value

Oracle’s filings report cloud services and license support, cloud license and on-premise license, hardware, and services. They do not publish a standalone audited income statement for Fusion AI agents or Oracle Health AI. Contract announcements, remaining performance obligations, and cloud infrastructure demand also should not be attributed entirely to Sicilia’s applications portfolio.

This is a recurring measurement problem in enterprise AI. A feature may improve renewal, protect an application seat, increase consumption, or support a larger cloud contract without appearing as separate AI revenue. Buyers should measure their own outcome: accepted transactions, cycle time, error rate, manual review, rework, and all-in cost.

Procurement questions

Oracle customers should ask which feature is generally available, which model processes the request, which region handles data, how customer content is retained, and whether prompts or outputs are used for training. They should map every tool action to an identity and approval rule. Logs, evaluations, and policy configuration should be exportable.

Sicilia’s strategic position is strong where Oracle already owns the transaction and data model. That same position increases switching costs and concentrates operational risk. Public evidence confirms the leadership structure and product direction. It does not establish broad autonomous operation, universal return, or error-free deployment. A bounded workflow with customer-controlled evidence remains the appropriate test.