Datadog CISO on Securing AI Agents at Scale | Deep Dives with a16z
a16z Deep Dives · 2026-08-11 · 22 min
https://www.youtube.com/watch?v=KSYnuCqCL4oVideo summary
Datadog CISO Emilio Escobar explains how 4,000 engineers use AI agents, while role-based MCP servers and an LLM judge help catch malicious code.
Datadog CISO Emilio Escobar tells a16z’s Joel De La Garza that Datadog moved from 50 Cursor licenses to AI adoption across nearly every employee, including more than 4,000 engineers. Rather than block tools, the company governs access: role-based MCP servers limit which business data employees can query, while sandboxing keeps agents away from stored credentials and provides short-lived tokens when needed. Datadog also built an LLM-based “judge” to assess code and AI skills for malicious intent, after finding harmful skills in marketplaces and detecting injected code in compromised packages. Escobar warns that agents can misuse permissions or “fix” a problem by disabling a database. He is less concerned about AI escaping sandboxes than the surge in vulnerabilities, weak processes for prioritizing them, and exaggerated trust in model-generated findings. His prescription: build security with engineers and avoid gatekeeping.
Chapters
- 0:00Deploying AI at Datadog: From 50 Cursor Licenses to Other Coding Agents
- 3:11AI Flattens the Org: 4,000 Engineers and a Sales Rep Querying Enterprise Data
- 5:19Role-Based MCP Servers and Sandboxing Agent Credentials at Datadog
- 7:34The Intent Judge: Datadog's LLM Catches Malicious Code and Agent Skills
- 10:27Reward Hacking: Datadog Checks Agents That Might Fix 4 a.m. Paging by Turning Off a Database
- 12:07The Helplessness Problem: CISOs Can Empower Developers Instead of Waiting for Vendors
- 14:05Security Engineers Will Become Real Engineers: Pay Parity, DevEx Collaboration, and Less Scanner Noise
- 18:12“AI Has Gone Wild”: Emilio Escobar on Access Rules, Vulnerability Volume, and Model Findings
What the full analysis includes
8 chapters · 40 key points · 8 section summaries · mind map
Preview of the first chapter · Deploying AI at Datadog: From 50 Cursor Licenses to Other Coding Agents
The remaining 37 key points, section summaries and mind map are in the full analysis. Analyze this video to see everything →
This is a Tier 1 public summary
Whether the chapter key points, section summaries and mind map are public is up to the person who shared it. Want the full analysis?Submit one yourself.
More from this channel
To Regulate AI Effectively, Focus on How It’s Useda16z Deep DivesMartin Casado argues AI laws should target harmful uses, warning regulatory uncertainty is pushing U.S. startups toward Chinese open-source models.
AI Copilots Are a Dead End. Here's What Actually Works | Kavak CEOa16z Deep DivesKavak CEO Carlos García Ottati says AI agents now handle 90–95% of customer interactions after a year of flat growth.
Are Tariffs Fueling the AI Race? | Deep Dives with a16za16z Deep DivesJacob Helberg says winning the AI race requires leading models, global adoption, secure supply chains, and rebuilding U.S. manufacturing.
Why Modern War Needs Intelligent Power Systems | Chariot Defense CEO on a16za16z Deep DivesChariot Defense’s M424 hybrid system delivered 36 hours of low-signature power as Army leaders push faster procurement and U.S. battery production.
Submarines and the Future of Defense Manufacturing | Hadrian CEO on a16za16z Deep DivesHadrian opened its 2.25-million-square-foot Alabama factory as the Navy faces 70 million submarine-production hours and a severe workforce gap.
Related analyses
Can Open Models Keep Up With OpenAI and Anthropic? | Nathan Lamberta16zNathan Lambert says China matches U.S. AI talent but lacks compute, while open models trail frontier labs on different timelines.
How To Stand Out In A World Of AI DesignsY CombinatorMagic Patterns CEO Alex Danilowicz demos MagicOS and critiques four startup sites, warning that confusing CTAs and scroll-jacking hurt conversions.
Cybersecurity in the Agentic Era | Deep Dives with a16za16z Deep Divesa16z guests say 50% of enterprise apps may be agentic by year-end, overwhelming security tools built for people and malware.
How Microsoft is Adapting to the AI Eraa16z Deep DivesMicrosoft Deputy CISO Aaron Zollman explains how OpenClaw can be enabled safely and AI could accelerate vulnerability patching.
用 AI 把自主權還給全盲的母親！當稀缺資源變成「算力競賽」，公平該怎麼重寫？｜Peggy【塞掐 Side Chat】E424INSIDE 硬塞的網路趨勢觀察NGO 工作者 Peggy 用 AI 助理讓盲人母親自主網購，也教公務員自動化行政；她提醒復康巴士等稀缺資源須重新設計公平規則。