
Already know some of this? A 2-minute test skips the lessons you know.
Take the placement testA free 13-lesson course built from Jon Krohn's 3h 34m of YouTube video. 3 days at 90 minutes a day, with study notes and a quiz in every lesson. A 2-minute placement test skips what you already know.
Taught by Jon Krohn on YouTube. Videos play in YouTube's own player, so the creator keeps every view.
Understand the workshop structure across four modules and learn the Anthropic definition of AI agents as programs where LLM outputs control the workflow with autonomy.
Evaluate the unprecedented opportunity for building agentic AI systems using market demand, LLM benchmark improvements (HLE), and cognitive task automation trends.
Identify ten enabling factors (LLM advances, cloud compute, data growth, etc.) and recognize high-impact agentic use cases like code generation, medical diagnosis, and scientific research automation.
Distinguish between predefined workflows and autonomous agents, and learn the actual mechanics of tool calling where LLMs respond with actions for code to execute, not direct system access.
Compare major agentic frameworks (no framework, MCP, OpenAI Agents SDK, CrewAI, LangGraph, AutoGen) and understand risks like unpredictable costs/outputs mitigated by guardrails and monitoring.
Meet co-presenter Ed Donovan and learn why OpenAI Agents SDK is his favorite framework: lightweight, minimal terminology (agents, handoffs, guardrails), and makes difficult things easy.
Master the three-step OpenAI Agents SDK pattern (create agent, use trace context manager, call runner.run) and understand the coding session approach: follow along, then practice later with GitHub repo.
Create a simple joke-telling agent using the OpenAI Agents SDK, then inspect execution traces to verify agent behavior and debug workflows.
Build a multi-agent deep research system using structured outputs (Pydantic) with a planner agent that designs search strategies, parallel search agents, and a writer agent that synthesizes reports.
Extend the deep research agent with a push notification tool via function decorators and wrap the workflow in a Gradio web interface for user-friendly interaction.
Learn best practices for structured outputs (Pydantic), debugging multi-agent systems (start small, monitor traces, guardrails), cost controls (API limits, cheap models), and visualization options.
Apply Anthropic's five workflow patterns: prompt chaining for fixed subtasks, routing for specialized LLMs, and parallelization for concurrent LLM execution with code coordinators/aggregators.
Implement advanced workflows where LLMs dynamically orchestrate subtasks (orchestrator-worker) and validate/refine outputs iteratively (evaluator-optimizer) for higher autonomy and quality.