Capital One CIO Outlines Agentic AI Adoption Steps

Forbes published an interview by Jeff Koyen with Mark Mathewson, executive vice president and divisional CIO for bank technology at Capital One, framing agentic AI as the next stage of enterprise AI that will take action rather than only advise. In the Forbes interview Mathewson said, "[Becoming an agentic enterprise] means building AI that understands its environment and can make decisions ... in a secure, active environment," and added that "Our history on the cloud has been foundational to preparing us for this next transition," according to the article. Mathewson told Forbes that "Data, which is such an important component of being an agentic enterprise, is now available in the cloud with access to all the variety of different compute services and models in a seamless, on-demand, elastic environment," per Forbes. The article frames these remarks as part of four strategic insights for enterprise adoption of agentic AI.

Industry-pattern observations: agentic systems move beyond single-turn LLM responses and require integrated stacks combining centralized data access, scalable compute, and runtime orchestration. Enterprises adopting agentic workflows typically need elastic provisioning for both model inference and stateful execution, stronger end-to-end observability, and production-grade grounding sources to limit hallucination risk.

public-cloud adoption and consolidated data platforms lower the operational friction for running agentic workloads at scale, because they provide on-demand compute, managed services, and standardised identity and networking primitives. For practitioners, operationalizing agentic AI raises familiar but amplified concerns around secure decisioning, access controls, audit trails, and testing of multi-step, stateful behaviors.

• •Adoption indicators: increased offering and uptake of managed orchestration layers, agent runtime platforms, and production telemetry focused on decision outcomes.
• •Security and compliance signals: new vendor features or frameworks for auditable decision logs, granular RBAC for autonomous actions, and industry guidance for liability and approval workflows.
• •Data-platform evolution: expanded investments in centralised feature stores, real-time data feeds, and cheaper nearline compute that enable stateful agents to act reliably.

Practical implications include prioritizing data accessibility, designing robust control surfaces for agent actions, and investing in observability and testing strategies that validate end-to-end agent behaviors before granting production permissions. These are general, sector-wide considerations and are not claims about Capital One's internal roadmap beyond what was reported in the Forbes interview.