Infrastructure & Ops Superstream Highlights

Why It Matters

Platforms become the critical control point for safely scaling AI: without robust platform governance and observability, rapid democratization of coding risks security incidents, compliance failures and operational chaos. Investing in platform hardening preserves velocity while preventing unsafe or unmanageable agent behavior.

Summary

Speakers warned that as AI agents proliferate, platform engineers must prioritize security, governance and observability over the flashy agent logic. They urged firms to build the “boring” infrastructure first—encryption, cross-instance coordination, telemetry, and strict isolation—because agents can interact unpredictably with external systems. With AI lowering the barrier to software creation, more non-engineers will ship tools, increasing the need for guardrails, evals, and platform-level controls. The panel argued companies should shift engineers from feature work to hardening platforms that make agent-driven software safe and durable.

Original Description

Agentic AI is moving faster than most platforms can safely absorb. At O'Reilly's recent Infrastructure & Ops Superstream, practitioners who are actually solving that problem shared what it really takes to operationalize AI agents inside organizations that can't afford to get it wrong.

Google Cloud’s Abdel Sghiouar broke down why traditional "secure by default" thinking breaks when agents can spin up browsers, execute code, and live for 24 hours, and why isolation is your first and best defense. Cockroach Labs’ Jordan Lewis underscored that the agent loop is the easy part, and the real work is the boring stuff: auth models, encryption, telemetry, and everything else that separates a toy from something a company can actually trust. Syntasso’s Daniel Bryant argued that AI is accelerating software creation faster than organizations can safely operationalize it, making platform investment more urgent than ever. Technology leader Sarah Wells pointedly noted that governance that creates friction will simply be coded around, especially now that anyone in marketing can vibe code their own tools. Thoughtworks’ Ben O'Mahony made the case for treating evals as a core part of your observability stack. And DORA’s Nathen Harvey closed things out by contending that now that everyone in the organization can build, every product engineer working on features right now should probably just stop and shore up the platform instead.

Watch the highlights now.

Follow O'Reilly on:

LinkedIn: https://www.linkedin.com/company/oreilly/

Facebook: http://facebook.com/OReilly

Instagram: https://www.instagram.com/oreillymedia

BlueSky: https://bsky.app/profile/oreilly.bsky.social

TikTok: https://www.tiktok.com/@oreillymedia