Keep pulling the thread on Katelyn Lesse & Angela Jiang.
Anthropic's platform roadmap is projected to evolve from a "knowledge layer" to an "execution layer" and subsequently to a "coordination layer" of abstractions.
When selecting verticals, Anthropic's product strategy favors "token hungry" industries like coding, where completing one AI-assisted task encourages further use.
Anthropic believes the most significant innovation in agentic systems will come from "meta harnesses" or "strategies" that allocate tokens to different jobs, such as execution, reflection, or advising a smaller model.
Anthropic may offer a model router product in the future, but it will be designed to route tasks only between different Claude models, not to external models.
A near-term focus for the Anthropic platform team is building tools for developers to compose "strategies" at the "coordination layer" of abstraction.
Anthropic's platform strategy is heavily indexed on the concept that a "token has a job," and the company aims to make it easier for developers to experiment with assigning different jobs to tokens.
Anthropic provides standards like SKILS and MCP to the developer ecosystem to help them achieve the best results with the Claude model.
Anthropic's platform team dogfoods new products internally while simultaneously offering early access to external customers to gather diverse feedback.
The Anthropic Platform team manages both the company's externally facing developer APIs and its internal product infrastructure.
Anthropic's platform strategy includes close integration with hyperscalers like AWS and Google to be physically and logically close to its business customers.
AI-native startups tend to use Anthropic's low-level primitives, whereas enterprises and other startups prefer higher-order, packaged offerings.
Anthropic's high-level product for its agentic "execution layer" is called Claude Managed Agents.