Software is increasingly creating work for other software. A human can set a goal, while agents plan, delegate to specialized subagents, review outputs, and combine results through many model calls.
The internet was built around human visitors. Products will increasingly need to support agents acting for people or other systems with clear identity, permissions, APIs, structured data, rate limits, observability, and audit trails.
The practical question for product teams is whether an agent can discover, understand, authenticate with, use, and verify the product without a human clicking every button.