Technology Now · Hewlett Packard Enterprise

What is an AI agent?

May 15, 2025·21 min·2 clips

Jimmy Whittaker reveals how AI agents force large language models to plan steps like humans, automating multi-step tasks.

Technology Now from Hewlett Packard Enterprise features a detailed explainer on agentic AI with HPE Chief Scientist Jimmy Whitaker. Whitaker begins with a concise LLM primer: large language models learn by masking words in training text and predicting what belongs in those slots, a process repeated across internet-scale data until the model infers the statistical structure of language. He describes hallucinations as the central current problem — LLMs confidently generate plausible-sounding outputs for things they cannot actually know, like recent events after their training cutoff. Agents address this by wrapping the LLM in a control loop: rather than generating a single answer, the model is given a goal, a list of tools it can call, and instructions to plan its next steps. After each tool call, the results are fed back into the model, and it plans again, iterating until the task is complete. Whitaker uses booking a Scotland flight as a tangible example requiring multiple sequential steps: generating candidate cities, looking up flights, comparing prices — tasks that an LLM alone could not accomplish. He mentions the Model Context Protocol (MCP) as an emerging standard for unifying how agents call and manage tools. He acknowledges limitations: agents can get stuck in ping-pong loops retrying the same failed approaches, and hallucination risks persist. On future directions, he predicts networks of specialized agents communicating with each other — speech-to-text, reasoning, text-to-speech — and frames agentic AI as likely to replace some jobs but more broadly requiring a new skill set, analogous to how the industrial revolution required new management and process knowledge. The Today I Learned segment covers a 2024 discovery that a 2004 magnetar flare — a neutron star with an intense magnetic field — created up to one-third of Earth's mass in heavy elements including gold and platinum, potentially accounting for 10% of all heavy elements in the galaxy. The This Week in History segment covers the 1980 WHO announcement of smallpox eradication, achieved through ring vaccination — vaccinating only the close contacts of confirmed cases rather than mass-vaccinating entire populations, a targeted strategy now also shown to reduce Ebola spread.

As heard by us

A clear, practical look at what agentic AI adds and what it still cannot solve.

Technology Now gives a clean look at agentic AI and the gap between today's systems and agents that can keep work moving. It works best when it stays grounded in specifics: agents looping through tools, prompts, and repeated context to reach a result, which makes them more…

Read the full review in PlayNext →

Why you'd press play

Want a straight answer to what agentic AI actually does?

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Technology Now