Claude Agent SDK [Full Workshop] — Thariq Shihipar, Anthropic | AI Engineer

The podcast explores the Claude Agent SDK, focusing on its architecture, capabilities, and practical applications in building AI agents. It emphasizes the SDK's foundation on Claude Code and its unique approach to agent design, highlighting the importance of Unix primitives like Bash and the file system. The discussion covers the advantages of code generation for non-coding tasks, security considerations, and the roles of tools, skills, and sub-agents in agent development. The podcast also addresses the trade-offs between workflows and agents, context management, and strategies for verifying agent work. Examples include building agents for spreadsheet analysis, cybersecurity, and even a Pokemon team, illustrating the SDK's versatility and potential.

Outlines

Part 1: Introduction, SDK Basics

Part 2: Architecture, Tools, Security

Part 3: The Power of Bash

Part 4: Agent Design, Loops, Verification

Part 5: Context Management, Skills

Part 6: Integration, Data, RBAC

Part 7: Specialized Agents, Tabular Data

Part 8: Advanced Problem Solving, Sub-Agents

Part 9: Prototyping, Case Studies, Future

Sign in to continue reading, translating and more.

Continue

Claude Agent SDK [Full Workshop] — Thariq Shihipar, Anthropic

AI Engineer

Part 1: Introduction, SDK Basics

00:21Introduction to Claude Agent SDK: Collaborative Exploration of Agent Building

Introduction to Claude Agent SDK: Collaborative Exploration of Agent Building

00:51The Evolution of AI Features: From Single LLMs to Autonomous Agents

The Evolution of AI Features: From Single LLMs to Autonomous Agents

03:30Claude Agent SDK: Addressing Redundancy in Agent Development

Claude Agent SDK: Addressing Redundancy in Agent Development

05:55Applications and Advantages of the Claude Agent SDK

Applications and Advantages of the Claude Agent SDK

Part 2: Architecture, Tools, Security

07:56The Anthropic Approach: Unix Primitives and Code Generation for Agents

The Anthropic Approach: Unix Primitives and Code Generation for Agents

09:09Agent Hosting Architecture and Initial Questions on the Agent SDK

Agent Hosting Architecture and Initial Questions on the Agent SDK

10:14Code Generation for Non-Coding: An Example of Weather API Integration

Code Generation for Non-Coding: An Example of Weather API Integration

11:26Workflows vs. Agents: GitHub and Slack Automations with the Claude Agent SDK

Workflows vs. Agents: GitHub and Slack Automations with the Claude Agent SDK

12:35Security and Guardrails: A Multi-Layered Approach to Agent Protection

Security and Guardrails: A Multi-Layered Approach to Agent Protection

Part 3: The Power of Bash

15:25The Power of Bash: Enabling Code Mode and Programmatic Tool Use

The Power of Bash: Enabling Code Mode and Programmatic Tool Use

17:25Non-Coding Agents: Using Bash for Email Analysis and Dynamic Information

Non-Coding Agents: Using Bash for Email Analysis and Dynamic Information

19:08Bash Tool Examples: Email and Video Meeting Agents

Bash Tool Examples: Email and Video Meeting Agents

Part 4: Agent Design, Loops, Verification

20:32Workflows and Agents: Flexibility with the Agent SDK

Workflows and Agents: Flexibility with the Agent SDK

21:34Designing an Agent Loop: Reading Transcripts and Key Components

Designing an Agent Loop: Reading Transcripts and Key Components

22:34Verifying Agent Work: Linting, Compiling, and Citing Sources

Verifying Agent Work: Linting, Compiling, and Citing Sources

24:16Planning and To-Do Lists: Enhancing Agent Execution with the Agent SDK

Planning and To-Do Lists: Enhancing Agent Execution with the Agent SDK

25:33Tools, Bash, and CodeGen: Choosing the Right Approach for Agent Tasks

Tools, Bash, and CodeGen: Choosing the Right Approach for Agent Tasks

27:50Strategic Tool Use: Atomic Actions, Composable Actions, and Dynamic Logic

Strategic Tool Use: Atomic Actions, Composable Actions, and Dynamic Logic

Part 5: Context Management, Skills

29:45Managing Long Outputs: Storing Results in the File System

Managing Long Outputs: Storing Results in the File System

30:37Skills for Bash: Enhancing Agent Capabilities with Context Loading

Skills for Bash: Enhancing Agent Capabilities with Context Loading

32:26Skill.md vs. Claude.md: Progressive Context Disclosure

Skill.md vs. Claude.md: Progressive Context Disclosure

34:07Model Evolution and Agent Code: Adapting to Advancing Capabilities

Model Evolution and Agent Code: Adapting to Advancing Capabilities

36:20Skills vs. APIs: Progressive Disclosure and File System Context

Skills vs. APIs: Progressive Disclosure and File System Context

Part 6: Integration, Data, RBAC

37:55Custom Bash Tools: File System Integration and Agent SDK

Custom Bash Tools: File System Integration and Agent SDK

39:19Agent SDK: React vs. jQuery and Performance Trade-Offs

Agent SDK: React vs. jQuery and Performance Trade-Offs

40:26Custom Bash Tools: File System Integration and Agent SDK

Custom Bash Tools: File System Integration and Agent SDK

41:55Agent SDK: Building Generic Chat Agents and Claude.AI Integration

Agent SDK: Building Generic Chat Agents and Claude.AI Integration

43:41Tools vs. Bash vs. Cogen: Accessing Databases and Guardrails

Tools vs. Bash vs. Cogen: Accessing Databases and Guardrails

46:48Role-Based Access Control: API Keys and Back-End Services

Role-Based Access Control: API Keys and Back-End Services

47:33Memory Tool: File System and Claude Agent SDK

Memory Tool: File System and Claude Agent SDK

48:08Agent Communication: Reusability and HTTP Requests

Agent Communication: Reusability and HTTP Requests

Part 7: Specialized Agents, Tabular Data

50:32Designing a Spreadsheet Agent: Search, Execution, and Linking

Designing a Spreadsheet Agent: Search, Execution, and Linking

51:43Spreadsheet Agent: Searching and Gathering Context

Spreadsheet Agent: Searching and Gathering Context

53:04Tabular Problems: Revenue in 2026 and Multi-Dimensional Data

Tabular Problems: Revenue in 2026 and Multi-Dimensional Data

54:49SQLite and CSV: Creative API Interfaces

SQLite and CSV: Creative API Interfaces

56:07Tool Ranking and Metadata: Enhancing Agent Performance