The Rundown Deep Dive: Why Agentic AI is the Next Phase of the AI Boom
20 snips
Mar 21, 2026 A breakdown of agentic AI and how autonomous agents differ from chatbots. Examples of current agent tools and live demos that can perform transactions. Why inference spending may overtake training and the hardware changes that requires. How big companies are already deploying agents and what that could mean for jobs, payments, and corporate capex.
AI Snips
Chapters
Transcript
Episode notes
Agentic AI Does Tasks Not Just Replies
- Agentic AI moves beyond prompts to autonomous task execution with tool use and multi-step planning.
- Zaid Admani explains agents can log into apps, call APIs, and complete jobs like booking meetings or filing reports with minimal supervision.
Inference Spending Will Overtake Training
- The industry shift from training to inference boosts demand for different infrastructure and far more compute in production.
- Gartner predicts inference capex will surpass training this year and by 2029 nearly double training spend, per Zaid Admani.
Agentic Workloads Need CPUs And More Tokens
- Agentic workflows use many more tokens and require CPUs alongside GPUs to orchestrate multi-step tasks and tool calls.
- Zaid notes Nvidia introduced the Vera CPU to handle inference and agentic workloads at GTC.