Jim Liddle, former Chief Innovation Officer at Nasuni focused on Data Intelligence and AI, founded Pract AI in January to bridge the gap between AI potential and implementation. He argues that AI buyers and managers lack the broad IT background knowledge of traditional IT professionals, partly because ChatGPT, the first generally available powerful Large Language Model, arrived only at the end of 2022. Liddle stresses that executives must learn about agentic AI capabilities and limitations, including GPU memory capacity constraints and work-arounds like KV Caching, to make reliable decisions on AI agent selection and deployment. He compares this to needing to understand electric car features to choose between an internal combustion engine vehicle and a battery-powered car. Liddle has produced explainers on KV Cache, context memory, and AI agent access control. He notes that AI agents, as digital employees working 24x7, require human-equivalent access controls, but their implementation must handle thousands of agents acting with microsecond latencies, where mistakes can expand the blast radius massively and instantly. He recommends giving an AI agent the least amount of access privilege needed, only for as long as needed, implying all agents must be known.