Treating AI agents as first-class infrastructure

AI agents are rapidly moving from side projects to production workloads, opening tickets, deploying code, querying databases, and interacting with internal systems. For platform engineers, this introduces a new class of non-human identity that does not fit existing IAM models. Agents need credentials, scoped access, auditability, and lifecycle management, yet many teams are solving this with API keys, service accounts, and ad hoc policy. This talk breaks down the emerging identity challenges of agentic systems and shares a practical framework for treating agents as first-class infrastructure. It covers patterns for strong agent identity, short-lived credentials, policy-based access, and auditability so platform teams can enable AI innovation without losing governance or control.