Enterprises are layering AI onto infrastructure that wasn't designed for it, with 57% prioritizing AI capability additions over rehosting or replacing systems, yet 72% having modernized less than half their app portfolio. Platform Engineering 1.0 focused on self-service infrastructure for human developers; 2.0 extends that to autonomous agents that provision resources, open pull requests, and act at machine speed. This requires platforms to expose APIs, policy, identity, governance, and cost controls consumable by both humans and machines, treat GPUs, model servers, MCP gateways, and vector databases as core primitives, and shift FinOps from after-the-fact reporting to real-time enforcement. Core principles like paved roads and platform-as-a-product remain unchanged; only the primary consumer shifts. This is the first piece in a three-part series.
Table of contents
About Prashanth ShenoyQuestions this post answers
What is platform engineering 2.0 and how does it differ from traditional platform engineering?
Platform engineering 2.0 extends the self-service, paved-road model of traditional platform engineering to support autonomous AI agents alongside human developers. It requires platforms to expose APIs, policy, identity, governance, and cost controls that both humans and machines can consume consistently. Core principles like paved roads, self-service, and platform-as-a-product stay the same; what changes is the primary consumer of the platform. Teams rethinking their platform for AI agents can track this shift in thinking on daily.dev.
Why doesn't traditional after-the-fact FinOps reporting work for AI agent workloads?
Autonomous agent workflows can consume resources non-linearly and incur unpredictable costs even when a run fails, so after-the-fact cost reporting arrives too late to prevent overspend. Cost governance instead needs to operate as a real-time constraint enforced at the API and policy layer, before or during resource consumption rather than only after it happens. Anyone building cost controls for agentic workloads can follow this FinOps shift on daily.dev.
What new infrastructure primitives does agentic AI add to a platform engineering stack?
Supporting AI agents means treating GPUs, large language model servers, MCP (Model Context Protocol) servers and gateways, and vector or data services as first-class platform primitives. These have scaling and cost characteristics fundamentally different from traditional compute and must be governed, authenticated, and scoped the same way as any other platform resource. Engineers evaluating MCP gateways and vector infrastructure can follow related coverage on daily.dev.