Head of API Engineering at Anthropic, leading the Claude Developer Platform encompassing APIs, SDKs, documentation, and console experiences. Previously Senior Director of Engineering at Betterment and Head of Engineering at Stripe. Expertise in platform development and building developer-facing products at scale.
Champion of Autonomous AI Agents
Katelyn Lesse has become a leading voice in building “unhobbled” autonomous agents, advocating for giving Claude computer access, external system connections, and code execution capabilities. Her work focuses on enabling agents to work effectively for hours on complex tasks without hitting context limitations.
Current Work
As Head of API Engineering at Anthropic, Katelyn leads the Claude Developer Platform evolution from basic API access to a comprehensive agentic harness:
- Model Context Protocol (MCP) - Open-source standard for connecting AI to external systems (“USB-C port for AI”)
- Memory System - Context management tools for long-running agents
- Context Editing - Manual window management to remove unnecessary content
- Code Execution - Native tool capabilities without external dependencies
- Tool Loop Framework - Minimal scaffold for tool calling in agent systems
Background
Previous Roles:
- Senior Director of Engineering at Betterment (cash management and business engineering teams)
- Head of Engineering for Core Connect at Stripe
- Engineering Manager of Managers for Stripe Capital
Notable Writing:
- A Functional Approach to Penny-Precise Allocation - Betterment Engineering Blog (2017)
- Medium - Engineering insights
Philosophy on Agent Development
Katelyn’s approach centers on balancing capability expansion with constraint management:
Core Agentic Stack:
- Tool Integration & Skills Framework - Structured capabilities for agents
- MCP - Composable connections to external systems
- Memory System - Intelligent context retrieval
- Context Editing - Token management and window optimization
- Code Execution - Native tool capabilities
Best Practices:
- Memory + Context Editing Combined = optimal context management
- “Giving Claude a Computer” - foundational metaphor for autonomous work
- Long-running Agents - techniques enable Claude to work for hours on complex tasks
Key Releases & Announcements
Claude Sonnet 4.5 - Latest model with significant improvements for agent development:
- Native code execution and context management capabilities
- Memory tools enabling hours-long operations
- Parallel tool execution
- Can maintain focus for 30+ hours on multi-step tasks
About Anthropic
Anthropic is an AI safety and research company building reliable, interpretable, and steerable AI systems. As a public benefit corporation, Anthropic balances investor returns with a mission to ensure transformative AI helps people and society flourish through frontier AI research and novel safety research from interpretability to RLHF to policy analysis.
Related Content
- Enabling Claude Code to Work More Autonomously
- Writing Tools for Agents
- Code Execution with MCP
- Unhobbling Claude: Anthropic’s Vision for Autonomous AI Agents - Video coverage
Conference Appearance
Event: AI Engineering Code Summit 2025 Date: November 20, 2025 Time: 8:49 AM Session: Evolving Claude APIs for Agents
Katelyn presented Anthropic’s latest agent capabilities, covering the core agentic system stack (tools, MCP, memory, context editing, code execution) and best practices for building powerful autonomous agents that can work for hours on complex tasks.