We went from Infrastructure as Hardware to Infrastructure as Code. The next leap? Infrastructure as Context — where AI agents don’t just execute infrastructure commands, they understand your infrastructure deeply enough to reason about it.

The context problem in platform engineering

Every platform team I’ve worked with has the same challenge: the infrastructure is complex, the documentation is always behind, and the tribal knowledge lives in people’s heads.

A senior engineer knows that “the checkout service is sensitive to latency because it has a synchronous dependency on the payment gateway, which has a 2-second timeout.” That context is nowhere in the code. It’s not in the runbook. It lives in experience.

What if AI agents could absorb that context?

From commands to understanding

Today’s infrastructure automation is command-driven: “scale this up,” “deploy that,” “restart this service.” The automation doesn’t understand why. It just does what it’s told.

AI agents with infrastructure context can operate differently. Instead of executing predefined runbooks, they can:

  • Reason about dependencies — “I shouldn’t restart this service because it would cascade to three downstream services during peak hours”
  • Understand intent — “the developer wants to deploy a new version, which means I need to check the canary metrics, not just push the container”
  • Learn from history — “the last time this alert fired, the root cause was a connection pool exhaustion, not the service itself”
  • Suggest, not just execute — “this Terraform change will increase your monthly cost by ~$2,400. Here’s an alternative that achieves the same goal for $800”

What “context” actually means

For an AI agent to be useful in infrastructure, it needs access to multiple layers of context:

  • Topology — what services exist, how they connect, what they depend on
  • State — current health, resource utilization, recent changes
  • History — past incidents, deployment patterns, known failure modes
  • Policy — compliance requirements, cost constraints, SLOs
  • Intent — what the operator is trying to achieve, not just what they’re asking for

This is where protocols like MCP become critical. They provide a standardized way for AI agents to access all of these context layers through a single interface.

What platform teams should build

If you’re building an internal developer platform today, start thinking about how your platform surfaces context, not just capabilities:

  • Make your service catalog machine-readable — dependencies, SLOs, ownership
  • Structure your incident history so AI agents can learn from it
  • Build your platform APIs to return not just data, but context about that data
  • Consider MCP servers for your key internal tools

The shift

Infrastructure as Code gave us reproducibility. Infrastructure as Context will give us intelligence. The platform teams that start building contextual foundations today will be the ones with the smartest AI-powered operations tomorrow.

The infrastructure isn’t changing. How we understand it is.