Building the enterprise environment for agentic AI (www.technologyreview.com)

Intel has announced key findings from its research on agentic AI, emphasizing its potential to revolutionize enterprise workflows beyond traditional tasks like chatbots. Agentic AI represents a holistic approach where automated software agents can execute complex business processes by integrating decision-making, task orchestration, and resilient data management across systems. Intel's experiments highlight that the true challenge extends beyond mere inference to encompass overall system performance, scalability, and observability. Enterprises should focus on metrics such as task success rate and task latency, which reflect the operational efficiency of agentic AI. One critical insight is the need for enterprises to plan based on agent density—measured as agents per virtual CPU—rather than simply counting agents. This metric serves as a vital indicator of system saturation and informs scaling strategies. Additionally, Intel provides a framework for deploying agentic AI systems that prioritizes scaling out (adding more systems) rather than scaling up (upgrading a single system), which can enhance performance, availability, and cost-effectiveness. By creating a reliable environment that manages costs, maintains governance, and effectively integrates agentic AI, businesses can leverage this technology to improve productivity and operational excellence.