On Cloud

On Cloud is a podcast produced by Deloitte about what happens inside large organizations when AI, cloud and newer technologies move faster than the systems meant to hold them. Gary Arora is the host. He introduces himself as chief architect for cloud and AI solutions at Deloitte, and his introductions name the pressure point up front: in one, architectures that cannot keep pace with AI; in another, an attack surface growing faster than teams can defend it. Then the guests take over. In the episode titled AI readiness: Bridging the gaps in enterprise architecture, a guest separates experimental projects, where upstream open source makes sense, from work already bound for production, where integration complexity should be retired as early as possible. Gary then raises a reversal. He says he spent much of the past decade helping enterprises move workloads and data into the cloud, and now sees some of those workloads heading back on-prem as companies explore running large or small language models there for data gravity, closer governance or cost predictability. The security conversations go into the same detail. On the cybersecurity episode, a guest argues that identity and access management has to come first, and that entitlements may need to move from roles toward attributes of the data once autonomous agents and sub-agents operate with narrow goals. Agents come back in a conversation about designing and governing them at scale. There, data quality splits into two questions, whether information is correct and whether it is current, illustrated by the gap between a tax return and a stock price. The same discussion lists newer threats such as prompt injection, decision loops and data leakage across agents. Healthcare keeps surfacing as the case where mistakes cost the most. Dr. Sanjeev Suratwala, who helps oversee AI governance at his hospital, describes governance built on clarity about who is responsible and how a model was trained, accountability that traces failures instead of blaming the algorithm, and enough agility to avoid blocking new tools. In his view, AI works best when humans stay in control. Ganesh Sitaraman, a managing director within Deloitte's AI, Data, and Engineering practice, also uses a hospital example: a virtual assistant triaging patient requests, which he uses to argue that AI systems must be explainable, fair and accountable across their lifecycle. Some episodes are labelled Knowledge Short and go to other Deloitte voices. Natasha Buckley, a senior research leader in emerging issues at the Deloitte Center for Integrated Research, hosts one on quantum computing with her colleague Diana Kearns-Monalato, making the case for a quantum roadmap now and using foresight analysis to sketch four visions of the future against three-to-five-year planning cycles. Across these episodes, accountability is the question that keeps returning: who owns a model, which agent can touch which data, and how a failure gets traced. Episodes close with the same call to keep innovating and keep leading, plus a Deloitte disclaimer that speakers' views are their own. The back catalog is large, so the easiest way in is by problem: agents, security, healthcare governance or quantum planning.