How to Safely Deploy AI Agents that Write and Manage your IaC

In this IaCConf 2026 session, Davlet Dzhakishev (founder of CloudGenie) shares two years of lessons from deploying AI agents that write and manage infrastructure as code in production. Davlet walks through the principles, guardrails, and practical patterns that make AI agents safe to run on real infrastructure. The session covers why infrastructure as code remains essential in an agentic world (LLMs are non-deterministic; IaC is your repeatability layer), why agents should never run deployments directly, and why teams need to think about four planes: agent runtime, credentials and permissions, change control through pull requests, and observability and audit trails. Davlet introduces OpenGenie, the open source agent runtime that solves common problems like sandboxing, durable sessions, and replayable history. Memorable takeaways: Treat agents like team members. They should not bypass your normal guardrails. Expect agents to hallucinate. Build the checks and validations around them. The engineering systems you already built (PR reviews, policy checks, audit trails) are exactly what agents need to run safely. You need to be a very good engineer and architect to set up agents that are actually usable in production. This session is essential viewing for platform engineers, DevOps leaders, security teams, and anyone evaluating AI agents for infrastructure work. Watch every session on demand:    • IaCConf 2026   Learn more about the IaCConf community: https://iacconf.com Follow IaCConf on LinkedIn for community updates and announcements about IaCConf 2027:   / iac-conf