Descriptions:
In this panel discussion from Anthropic, Claude platform team members Katelyn and Angela lay out how agentic infrastructure has evolved over the past six months and where it is heading. The conversation covers managed agents, agent identity, MCP server composition, and the security barriers that are slowing enterprise adoption — all from the perspective of the team building the underlying platform.
A recurring theme is the shift from deterministic, step-by-step orchestration to outcome-based agent operation. The speakers describe a model where an agent receives a goal, negotiates the permissions it needs upfront (“I need A, B, C — not D”), executes autonomously, and generates an auditable service-account trail. This vision of agent identity as a first-class concept, separate from the human user identity, is positioned as the next major infrastructure primitive. MCP servers are highlighted as a clean abstraction for exposing one agent’s capabilities to another, enabling multi-agent architectures without tight coupling.
Real-world examples anchor the discussion: a manufacturing company using agents to encode institutional knowledge from retiring workers, and Shopify’s end-to-end development platform “River” as an example of organizations that have raised the ceiling on what agents can accomplish. Security, compliance assumptions from legacy IT models, and the absence of robust evaluation frameworks are identified as the three primary barriers preventing broader enterprise adoption of what Anthropic’s platform makes technically possible.
📺 Source: Claude · Published July 10, 2026
🏷️ Format: Keynote Launch
