The Boring Layer

AWS dropped a knowledge graph at Summit New York yesterday. Not a model. A managed graph that maps where an organization’s data actually lives so agents can stop guessing.

They called it AWS Context. It learns which data sources are authoritative, which paths work, which business rules apply. Every agent benefits from what one query discovers. The strategic admission buried inside the launch: the model was never the bottleneck. The wiring was.

The numbers back it up. AgentCore tasks grew 15x in six months. Southwest Airlines is restructuring its entire dev org around agentic tooling — 2,700 developers. A fintech startup replaced a 12-month roadmap with one engineer and eight weeks. None of this is a pilot.

Every major provider landed here independently. Google with Vertex Agent Builder. Anthropic with MCP and tool use. AWS with Context. The consensus is quiet but total: intelligence without operational plumbing produces expensive demos, not deployed systems.

Most AI conversations inside companies still orbit capability. Which model scores highest. What it can do in a sandbox. Nobody asks whether the agent knows where the data lives, which source is truth, what rules apply. That is an operations question, not a technology question. And it is the only one that matters once you move past the demo.