# Anthropic Platform: Building Blocks for AI _Anthropic's Katelyn Lesse and Angela Jiang discuss building a flexible AI platform with layers of knowledge, execution, and coordination._ **Updated:** 2026-08-22 **Published:** 2026-07-14 **Source:** https://www.startuphub.ai/ai-news/artificial-intelligence/2026/anthropic-platform-building-blocks-for-ai --- In a recent interview, Katelyn Lesse and Angela Jiang, key figures from Anthropic's platform team, shared insights into their strategy for building and evolving the developer platform for Claude, Anthropic's advanced AI model. Anthropic PlatformCoreexternal API for developers and internal product infrastructure for Anthropic's appsFrom the article 8 mentionsIn a recent interview, Katelyn Lesse and Angela Jiang, key figures from Anthropic's platform team, shared insights into their strategy for building and evolving the developer platform for Claude, Anthropic's advanced AI model.Two North StarsContextmaximize leverage for Anthropic teams and empower external builders with Claude toolsFrom the article 2 mentionsThe team operates with two distinct "north stars." Internally, the goal is to provide maximum leverage for Anthropic's teams to ship AGI-pillar products quickly and reliably.Layered AI AbstractionCoreplatform built with layers of knowledge, execution, and coordination for flexibilityFrom the article 2 mentionsThe discussion delved into the layered cake of abstractions Anthropic is building on top of its core AI models.Speed & EmpowermentEffectship AGI-pillar products quickly and reliably, enabling any builder to createFrom the articleThe team is particularly excited about innovations in "strategies" or meta-harnesses, where tokens can be allocated to diverse tasks like reflection, advice, or self-correction, optimizing for cost and speed.Evolving Form FactorsContextadapting from chatbots to more complex AI agents as capabilities growFrom the articleAnthropic's platform is designed to be adaptable to these changing form factors, providing robust tools that allow users to discover and implement them.Open EcosystemOutcomebuilding an open platform, not a walled garden, fostering broad innovationFrom the article 2 mentionsA core philosophy for Anthropic's platform team is fostering an open ecosystem.informsFuture StrategiesContextfocus on token budgeting and strategic evolution of the platform's capabilitiesFrom the article 3 mentionsThe goal is to compose these into sophisticated, orchestrated strategies. ## Anthropic Platform: More Than Just APIs Lesse explained that the Anthropic platform serves a dual purpose: it's both the externally facing API that developers build upon to access Claude's intelligence, and the internal product infrastructure that Anthropic's own applications utilize. This dual focus ensures a consistent and robust foundation for all users. ## Two North Stars: Speed and Builder Empowerment The team operates with two distinct "north stars." Internally, the goal is to provide maximum leverage for Anthropic's teams to ship AGI-pillar products quickly and reliably. Externally, the focus is on empowering any builder with the tools to work with Claude, enabling them to create whatever they envision. Jiang elaborated on the external north star, emphasizing the desire to be wherever the business is and to bring the platform "really, really close to that business." This close integration with hyperscalers like AWS and Google Cloud is crucial for enabling developers to express their product ideas, almost as if they were building custom software. ## The Layered Approach to AI Abstraction The discussion delved into the layered cake of abstractions Anthropic is building on top of its core AI models. At the base is "knowledge," which encompasses understanding the model itself and the data it needs. Primitives here include the messages API, tool usage, and context management. The next layer is "execution," where the focus shifts to getting Claude to perform work, not just provide information. This involves managing infrastructure for tasks like spawning sandboxes with proper governance and security, or handling storage for long-running sessions. Claude Managed Agents is a key product in this layer. The final, and perhaps most forward-looking, layer is "coordination." This layer deals with strategies, acting as a meta-harness for orchestrating different AI tokens or agents with specialized jobs, advising, executing, reflecting, or even advising smaller models. The goal is to compose these into sophisticated, orchestrated strategies. ## From Chatbots to Agents: Adapting to Evolving Form Factors Lesse highlighted the rapid evolution of AI use cases. "Two years ago, we were all like, everything's chat. Now everyone's like, forget chat and just like agents," she noted. Anthropic's platform is designed to be adaptable to these changing form factors, providing robust tools that allow users to discover and implement them. ## Building an Open Ecosystem, Not a Walled Garden A core philosophy for Anthropic's platform team is fostering an open ecosystem. They are not precious about where these components run, as evidenced by the launch of self-hosted sandboxes and partnerships with companies like Modal, Vercel, and Cloudflare. The emphasis is on the architecture and interfaces that enable powerful, reliable, and scalable agentic systems. The team is also focused on standards, such as MCP and skills, to ensure interoperability across the ecosystem. Beyond technical interoperability, they are also exploring standards around safety and preventing negative AI behaviors. ## The Future: Strategies and Token Budgeting Looking ahead, Anthropic plans to move further up the abstraction layers, from knowledge to execution, and then to coordination. The team is particularly excited about innovations in "strategies" or meta-harnesses, where tokens can be allocated to diverse tasks like reflection, advice, or self-correction, optimizing for cost and speed. They also noted the industry trend towards "token rationalization," where companies are moving beyond simply maximizing token usage to strategically employing them for efficiency and impact. This involves routing tasks to the most appropriate models, whether they are large, powerful ones or smaller, more cost-effective alternatives. --- Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory. © StartupHub.ai. All rights reserved. You may not republish this article in full without a license. Search engines and AI research tools may crawl and summarize for reference. Bulk reproduction or model training on this content requires a license. See https://www.startuphub.ai/terms.