How to build your AI agent

In this episode of The Times Tech Podcast, hosts Danny Fortson and Katie Prescott explore the rise of AI agents, focusing on how to build one using OpenClaw, an open-source AI agent framework that gained viral momentum in early 2026. They walk through the hands-on process of setting up their own AI agent, 'Norman,' on a Raspberry Pi, detailing the technical hurdles, security risks, and cost implications—especially around electricity and token usage. The episode highlights the shift from chatbots to 'action engines' capable of autonomously performing tasks like drafting documents, managing emails, and interacting with business systems. A key discussion centers on the future of the internet, with guest Aaron Levy, CEO of Box, arguing that we're entering a new era where the internet must be redesigned for AI agents, not just humans. He envisions a world of microtransactions, programmatic interfaces, and shadow systems that agents use to operate efficiently. Despite concerns about job displacement and security, Levy remains optimistic, believing AI will democratize engineering and innovation across industries, enabling non-tech companies to build powerful software for the first time. The episode concludes with a nuanced take on AI’s impact: not as a job killer, but as a force that will reshape work, demand new skills, and unlock unprecedented productivity. The episode delivers actionable insights: start small with a Raspberry Pi, prioritize security and data privacy, use cloud-hosted models to avoid local compute strain, and design workflows that leverage AI agents for repetitive tasks. It also urges businesses to reinvest savings from automation into innovation, not just cost-cutting. The hosts emphasize that while AI agents are powerful, they are tools—dependent on human oversight, ethical design, and strategic deployment. The overall tone is cautiously optimistic, balancing excitement about AI’s potential with sober awareness of its risks and complexities.