From AI Talk to AI Action: What Enterprises Must Solve to Deploy Autonomous Systems

Over the past three years, enterprise discussions around artificial intelligence have centered on generative capabilities and model outputs. Looking ahead, the focus will pivot toward a more complex question: what degree of independent action should organizations permit their AI systems to take? Solving this challenge requires far more than simply deploying better models.

On October 28, the Bloomberg Tech Event: Building the Agentic Enterprise, sponsored by Hewlett Packard Enterprise Co., will convene in Boston to address this shift. The gathering brings together operators who must live with autonomous AI systems in production, including chief information officers from Mercedes-Benz North America and other enterprises; chief security officers representing Booking Holdings and Carnival Corp.; Estée Lauder Cos.' head of data, analytics and AI; and James Robertson, HPE's global field chief technology officer for networking.

The presence of enterprise practitioners matters significantly. While vendors can promote autonomous operations, it falls to the operators themselves to determine whether, when and how to activate such capabilities.

The event follows HPE's Networking Investor Day, where leadership positioned the network as having evolved from basic infrastructure into a strategic asset for AI deployment. According to Rami Rahim, executive vice president, president and general manager of HPE Networking, "AI is reshaping the technology stack, making the network more strategic." While investors interpreted this as a growth opportunity, enterprise technology leaders should recognize it as a fundamental shift in operating models—the infrastructure supporting AI increasingly must manage itself.

Five priorities for enterprise leaders

The question has changed from insight to action

Generative AI has made data-driven recommendations commonplace. Most organizations now face a surplus of insights they cannot act upon. The central question has shifted: rather than asking whether AI can produce recommendations, enterprises must ask whether they can trust AI systems to execute decisions without requiring human sign-off at every stage.

A poorly drafted email from an AI system causes embarrassment. An AI agent that modifies network policies, redirects traffic or isolates a device based on faulty reasoning can paralyze operations. The critical challenge lies in determining which actions warrant full automation, which demand human review and which should remain manual processes.

Leading organizations typically do not treat autonomous operations as a binary choice. Instead, they allow systems to gradually expand their autonomy, one category of action at a time, building confidence through demonstrated results.

Uptime is the wrong scorecard

At the investor day, Sujai Hajela, HPE's executive vice president and general manager of campus and branch, articulated a principle central to the company's self-driving network approach: "Up is not the same as good." This insight applies directly to autonomous operations. A network can maintain full availability while simultaneously delivering degraded video calls or starving AI workloads of resources.

Transitioning from infrastructure-centric metrics to user and business outcome measurement is essential for autonomous systems. A system cannot resolve problems independently if it cannot recognize what problems look like from an end-user perspective.

ServiceNow Inc.'s Sajeev Nair, senior director of digital core services, provided a compelling example at the investor day. His organization had deployed numerous point products that "creating another fragmentation in your telemetry that is already fragmented." Following consolidation, manual work that previously consumed 2,000 to 3,000 hours annually dropped below 60 hours.

Enterprise practitioners should define success around outcomes that drive business value: engineering hours recovered, incidents prevented before reaching support queues and faster deployment of new services—not headcount reduction.

The network is becoming AI infrastructure, not just connectivity

Rahim emphasized a point every CIO investing in AI should internalize: organizations can allocate billions toward graphics processing units, yet if the network cannot satisfy performance requirements, "those resources are not fully utilized."

AE Natarajan, who leads HPE's routing business, outlined why AI traffic presents distinct challenges. Bidirectional traffic becomes more symmetrical, AI workloads resist caching techniques effective for video and agents generate continuous traffic patterns. As Natarajan noted, "Digital users don't fall asleep."

When AI agents constitute a new category of network user, the network must identify them, enforce access policies and monitor their behavior. This transforms the network into an identity and policy enforcement layer for AI, extending beyond its traditional role as a transport mechanism. This convergence creates overlap between infrastructure, security and governance budgets.

Security leaders will set the pace of autonomy

The event's speaker roster reveals a telling pattern: at a networking-focused conference, three speakers represent security leadership. In most enterprises, the chief information security officer, rather than the CIO, will ultimately control the speed at which autonomous operations expand.

Security executives confront a dual challenge. First, AI agents require connectivity to systems, data and each other, creating a growing population of nonhuman identities that conventional zero-trust architectures were not designed to accommodate. Second, adversaries now possess AI capabilities as well. Anthropic's Claude Mythos has enabled vetted organizations to discover more than 10,000 high- or critical-severity vulnerabilities, and this capability is expected to proliferate. Aging, unsupported network equipment now faces direct risk.

Conversations with security and network professionals reveal that the "Mythos moment" prompted many to reassess their infrastructure modernization strategies. Substantial portions of network infrastructure exceed a decade in age. These legacy switches and routers lack the capabilities required to defend against AI-generated threats.

David Hughes, HPE's senior vice president and general manager of SASE and security, argued at investor day that "the network as a sensor" can identify threats and that the network itself can contain them, extending zero trust principles to agents and devices. The question for CISOs becomes whether they view autonomy as a risk to manage or a defensive capability they require.

The answer is both. Human-paced security operations cannot match machine-speed attacks. The trust framework enterprises construct for autonomous operations will simultaneously determine their defensive effectiveness.

The blockers are mostly human and operational

When practitioners identify obstacles to agentic AI deployment, they rarely cite model limitations. Instead, they point to fragmented data, aging infrastructure, ambiguous accountability, workforce capability gaps, cost considerations and, fundamentally, trust.

HPE characterized trust in self-driving networks as evidence-based, with systems demonstrating their recommendations and reasoning before acting independently. This represents sound thinking, yet it demands clean telemetry—a resource many organizations lack. Rahim cautioned that when vendors promote self-driving capabilities, "you really have to look under the covers to see how much is actually there."

The event will also explore tokenomics, with Bloomberg Intelligence software analyst Niraj Patel participating. Return-on-investment discussions frequently become the sticking point where agentic AI initiatives stall. Leadership must demonstrate that autonomy reduces expenses or generates new revenue streams, not merely that it functions in controlled trials.

A roadmap for enterprise leaders

CIOs and CTOs transitioning from AI experiments to autonomous operations should consider these four steps:

• Establish autonomy tiers by categorizing which actions AI can execute independently, which require human authorization and which must remain manual. Reassess this classification quarterly and adjust based on operational performance.
• Prioritize data consolidation. Autonomous operations depend on reliable, comprehensive telemetry across the entire system. Consolidating tools and control planes, as ServiceNow accomplished, typically represents the most efficient path to achieving this consistency.
• Apply zero trust principles to AI agents. Implement identity-aware access controls, least-privilege enforcement and continuous monitoring of agent activities, enforcing these policies across the network infrastructure rather than confining them to individual applications.
• Decommission unsupported systems. Catalog infrastructure that no longer receives patches and replace it before AI-driven attacks target these vulnerabilities.

The path forward

The transition to autonomous operations will not be determined by the most advanced models or the largest GPU deployments. Success will belong to enterprises that establish the trust, data infrastructure and governance frameworks enabling AI to act securely. This explains why the event's composition matters: it assembles individuals accountable for availability, security and business performance alongside technology vendors on a shared platform.

HPE's Networking Investor Day served as an introduction to AI-driven autonomous systems, demonstrating that the company's self-driving network has progressed beyond conceptual presentations. The Bloomberg event represents the substantive exploration that will determine whether CIOs and CISOs can align on autonomous operations.