Contrary to the perception that agentic AI is nascent, research from Signal 65 shows massive adoption in enterprises. This rapid uptake is a bottom-up movement, with employees demanding access to the same AI tools they use personally for work purposes.
To navigate millions of public models, Dell built a Model Evaluation Protocol (MEP). This system uses AI agents to test new models across different hardware, plotting performance vs. intelligence on a Pareto curve to provide tailored, data-driven customer recommendations.
Agentic systems increase throughput by delegating tasks to numerous sub-agents. However, this parallelism comes at a cost. The total token consumption often increases because the sub-agents may not operate with maximum token efficiency, expanding the total workload.
The most advanced AI models are not universally superior; their capabilities form a "jagged frontier." This means organizations can often use more economical, locally-run open-weight models for tasks where they are "good enough," reserving expensive frontier models for specialized needs.
Unlike chatbots that simply respond to queries, agentic AI systems can interact with external tools, access data from various sources, and perform actions on a user's computer. This capability to operate outside its immediate interface is the key differentiator.
IT leaders face immense pressure from both senior leadership, who want to leverage AI for strategic advantage, and employees, who want productivity tools. This creates a difficult balancing act with their core responsibilities of ensuring system security, reliability, and managing tight budgets.
A one-size-fits-all approach to LLMs is inefficient. Power users develop rules of thumb for model selection based on the task's requirements. For instance, Grok is preferred for quick responses, while a more persistent model like Codex is used for complex, thorough tasks.
For teams with high-volume AI usage, the recurring cost of cloud-based, pay-per-token models can be enormous. Investing in on-premise hardware offers significant cost-avoidance, with systems achieving break-even in months and generating millions in equivalent value over their lifetime.
Unlike typical IT assets that depreciate, AI-capable hardware inverts this trend. As more efficient and powerful models are released, the same physical machine can tackle increasingly complex problems, making its utility and value grow throughout its lifecycle.
AI is democratizing software development. Non-technical employees can now create automated workflows by describing needs in plain English. This "citizen development" means code generation—and its associated token costs—is no longer confined to the engineering department but is happening across all business functions.