Dell Executive Warns Businesses About Costly AI Agents Without Proper Infrastructure Planning

Dell Technologies Chief Operating Officer Jeff Clarke has warned businesses that poorly planned artificial intelligence deployments could create “silent and expensive” AI systems that increase operational complexity, cloud spending, and security risks.

Speaking about the future of enterprise AI, Clarke outlined five major design principles companies should follow as they adopt autonomous AI agents and large-scale AI workflows across their organizations.

According to Clarke, many enterprises are rushing to implement AI tools without fully preparing their infrastructure, data systems, or operational controls for the long-term demands of agentic AI systems.

1. Bring AI to the Data Instead of Moving Data to AI

Clarke’s first recommendation focused on enterprise data management. He argued that most organizations still operate with fragmented data spread across multiple disconnected systems, making it difficult for AI agents to function effectively.

A major challenge is that a large percentage of enterprise information remains unstructured, meaning AI systems struggle to access and use it efficiently.

Instead of transferring huge amounts of corporate data into external cloud platforms, Clarke advised businesses to deploy AI models closer to where their data already resides. This approach can reduce security risks, improve performance, and lower infrastructure costs.

He emphasized that future AI-native enterprises will require real-time connected data systems capable of supporting autonomous agents at scale.

2. Prepare for Massive AI Inference Demands

Clarke also highlighted the rapid rise of AI inference workloads — the stage where AI systems actively generate outputs, make decisions, and perform tasks after training is complete.

Unlike simple chatbots, advanced AI agents often perform complex multi-step reasoning tasks involving multiple models and workflows simultaneously.

According to Clarke, these workloads are becoming dramatically more compute-intensive than traditional AI systems, forcing enterprises to rethink how they design computing infrastructure.

He stressed that businesses must build systems capable of supporting both AI training and high-volume inference operations if they want to scale autonomous AI tools effectively.

3. Track Every AI Action With Full Transparency

Security and accountability were another major focus of Clarke’s recommendations.

He warned that AI agents increasingly interact directly with critical enterprise systems such as customer databases, financial platforms, and operational software. Because these systems can autonomously trigger actions, companies must implement strong tracking and auditing capabilities.

Clarke argued that every AI-generated action should leave behind a traceable “receipt” explaining what was done, why it happened, and how it can be reversed if errors occur.

This level of visibility, he said, is essential for building trust in AI systems that operate independently across business operations.

4. Integrate the Entire Enterprise Software Stack

Another major risk involves fragmented software environments.

Clarke explained that AI agents need to coordinate tasks across multiple enterprise systems including CRM platforms, ERP software, databases, analytics tools, and workflow systems.

Without deep API integration and orchestration capabilities, businesses could end up with isolated AI agents that fail to communicate effectively and become extremely costly to maintain.

He warned that disconnected AI deployments could create operational inefficiencies and financial concerns that executives may later struggle to justify to company leadership and boards.

5. Control AI Costs Through Smart Token Routing

Finally, Clarke urged businesses to carefully manage the economics of AI usage, particularly as AI workloads continue growing rapidly.

He emphasized that not every task requires the most advanced and expensive AI model available. Instead, companies should intelligently route workloads to the most efficient infrastructure depending on the complexity of the task.

This concept, often referred to as “token routing,” involves deciding whether tasks should run on local systems, edge devices, private infrastructure, or cloud-based frontier AI models.

According to Clarke, organizations that fail to optimize AI routing strategies could face rapidly escalating cloud costs as AI adoption scales.

AI Adoption Enters a New Enterprise Phase

Clarke’s comments reflect broader concerns emerging across the technology industry as enterprises move from AI experimentation into large-scale deployment.

Many businesses have successfully tested AI systems in pilot projects but are now discovering the operational, financial, and governance challenges involved in scaling autonomous AI across entire organizations.