Jeffrey Aven

Platform engineering built its playbook for human consumers: portals, golden paths, self-service templates, and review gates. AI agents break most of those assumptions — they do not click through portals, they do not wait for reviews, and they consume capability at machine speed. This session looks at what changes when agents become first-class consumers of your internal platform, drawn from production experience building an agent-facing control plane across multiple cloud providers. We cover what agents actually need from a platform: queryable live state rather than dashboards, machine-enforceable policy rather than human review gates, and discoverable capability rather than documentation. We look at interface design for agent consumers — why exposing platform capability as a uniform query and provisioning surface beat a catalogue of bespoke tools, and what that did to context size and reliability — and at guardrails as platform features: read-only defaults, query-before-mutation, and policy gates enforced at the interface. Includes a live demo of an agent consuming platform capability end to end, from natural language question to policy-gated change, and an honest account of what we got wrong first. All patterns shown are tool-agnostic, demonstrated with open source software.