S01 - E06 - Demystifying AI agents_ Beyond the buzzword

Production incidents are exploding as AI accelerates how fast code ships, and the people left holding the pager are paying the price. In this episode, we sit down with Omri Sass, Senior Director of Product Management at Datadog, to talk about why firefighting is quietly burning out engineering teams, and how AI is starting to change that.This episode is brought to you by Datadog. As AI adds new agents, new tools, and new complexity across your stack, Datadog gives teams real time visibility into infrastructure, applications, and AI systems, so nothing breaks without you knowing why. Start your free trial at datadoghq.com.

Your AI agents might be talking to tools — and each other — in ways you never approved. Nate Bittinger, VP of Engineering at Docker, unpacks the fragmented, fast-moving world of AI tooling and the very real risks of prompt injection and tool poisoning. His fix: give every agent its own container, and back it with a centralized enforcement layer like an MCP gateway to catch and block threats without slowing teams down. He closes with an open Q&A on what it actually takes to bring Docker's approach into your existing stack.

Before an autonomous vehicle ever drives a real mile, Waymo has already driven it millions of times — in simulation. Smitha Shyam, Senior Director of Engineering at Waymo, shares how the company builds virtual worlds to stress-test the Waymo Driver against scenarios too rare or risky to wait for on real roads. With over 100 million driverless miles logged, she walks through the safety framework and sensor technology behind Waymo's push to become the world's most trusted driver — and what it takes to adapt autonomous driving across different countries and road conditions.

What actually makes something an "AI agent" — and when should you skip building one? Deepak Pai, Principal Scientist at Adobe, cuts through the hype to explain the real building blocks: LLMs as the "brain," paired with function calls, APIs, and the Model Context Protocol to let them perceive and act. He offers a clear-eyed take on when agents are worth the complexity (open-ended, multistep tasks) and when they're not (simple, high-stakes, or predictable work) — plus an honest look at where agents still fall short today. Q&A covers tooling choices and when to trust an agent versus your own expertise.

Jailbreaking. Prompt injection. Attacks your traditional security stack was never built to catch. Ashwin Kesireddy, VP of Product & GTM for AI Security at Zscaler, breaks down the three pillars enterprises need to secure AI at scale: visibility into every AI app and agent in use, guardrails on how employees interact with generative AI, and controls for the AI systems teams build in-house. He also demos Zscaler's red-teaming approach to stress-testing agents, and shows how automated remediation and compliance mapping close the gap traditional tools leave open. A must-listen for anyone responsible for AI risk.

As enterprises race to deploy AI agents at scale, how do you keep innovation and security from pulling in opposite directions? In this executive summit panel, Kal De (SVP, Product & Engineering, Nokia), Tejasvi Devaru (VP, IT Customer Experience Systems, Zendesk), and Hira Dangol (AI Executive & Advisor) unpack what it really takes to govern AI safely — from identity management for autonomous agents to building security frameworks for systems that no longer behave predictably. Expect a candid look at AI security gateways, policy frameworks, and monitoring strategies that let teams innovate without losing control. If you're building or buying AI agents, this conversation is your governance crash course.

Karanveer Anand leads technical program management for software reliability at Google, and in this episode, he breaks down what it actually takes to keep AI systems trustworthy at scale.