Descriptions:
Thariq Shihipar, an engineer at Anthropic, leads a two-hour hands-on workshop introducing the Claude Agent SDK—the open framework that underpins Claude Code and is now available for developers to build their own autonomous agents. The session covers the full arc from conceptual foundations to live prototyping.
Shihipar explains how Anthropic’s agent architecture evolved from rebuilding the same primitives repeatedly across internal projects: a tool harness, a core prompting layer, file-system access as a form of context engineering, and a looping execution model. The Claude Agent SDK packages all of this, including support for subagents, web search, memory, compaction, hooks, and skills (reusable instruction sets triggered automatically or via slash commands). A significant portion of the workshop addresses the practical art of tool design—when to expose database access directly via bash versus wrapping it in a structured tool, how to scope API keys for agent safety, and how role-based access controls fit into agentic workflows.
The live coding section demonstrates building an agent loop from scratch, with Shihipar candidly working through design decisions in real time rather than a polished canned demo. Developers evaluating agent frameworks, anyone moving from single-LLM features to autonomous workflows, or teams building on top of Anthropic’s APIs will find this workshop a thorough, authoritative starting point.
📺 Source: AI Engineer · Published January 05, 2026
🏷️ Format: Hands On Build
