Anthropic, OpenAI, Google, and Microsoft agree that the harness is the product. They disagree on the price.

Anthropic, OpenAI, Google, and Microsoft are converging on enterprise AI agent infrastructure as the key product layer.

The focus is shifting from the model itself to the agent control layer: runtime, tool orchestration, observability, and long-running sessions.

They are taking different commercialization paths: Anthropic charges separately for managed runtime, OpenAI open-sources the harness and bills only for model and tool use.

Google and Microsoft are bundling similar agent capabilities into broader offerings, aiming for easier enterprise adoption.

The market for production agent infrastructure is forming quickly, and vendor pricing choices will shape how enterprises build and buy agents.