Anthropic's Katelyn Lesse & Angela Jiang: Building an Ecosystem, not a Walled Garden

Anthropic’s developer platform architecture organizes AI capabilities into three distinct layers: knowledge, execution, and coordination. The knowledge layer focuses on model design and parameter standardization, while the execution layer provides managed infrastructure for agentic tasks. The emerging coordination layer utilizes "strategies"—or meta-harnesses—to orchestrate token usage, allowing developers to assign specific roles like advising or executing to different model calls. This modular approach enables builders to optimize for cost and performance while maintaining flexibility. By dogfooding these primitives internally, the team ensures that platform tools remain robust for both internal product development and external enterprise applications. Future innovation centers on standardizing interoperability through specifications like MCP and refining agentic workflows to handle complex, long-running tasks without requiring human intervention.