Back to Courses
DevelopmentAdvanced76 lessons50–65 hours
AI Agents from Scratch
Build autonomous AI agents that can browse the web, write code, manage tasks, and work 24/7.
What You'll Learn
Understand agent architecture: perception, reasoning, planning, action, memory
Design and build tools with proper schemas and error handling
Implement memory systems: short-term, long-term, episodic
Build agents with LangGraph, CrewAI, AutoGen, and Claude Agent SDK
Create 5 complete agent projects from research to code review
Deploy agents to production with cost management and observability
Implement safety guardrails and evaluation frameworks
Build multi-agent systems with real coordination patterns
Outcomes
- Build production-grade agents with LangGraph, CrewAI, AutoGen, and Claude Agent SDK
- Design agent memory, planning, and tool-use architectures from scratch
- Deploy multi-agent systems that handle real business workflows
- Implement safety guardrails, evaluation, and monitoring for agents
Prerequisites
- -Python proficiency
- -Experience with at least one AI API
- -Understanding of async programming
Projects You'll Build
- Personal research agent with web browsing
- Customer service multi-agent system
- Code generation and review agent
- Multi-agent business automation system
Course Curriculum
Module 1: What Are AI Agents? (Foundations)
- 1.1From chatbots to agents — the evolution of AI applications
- 1.2Agent anatomy: perception, reasoning, planning, action, memory
- 1.3The agent loop: observe, think, act, observe
- 1.4Types of agents: reactive, deliberative, hybrid, autonomous
- 1.5The agent landscape: frameworks, platforms, build-vs-buy
Module 2: Tool Use Deep Dive
- 2.1Why tools make agents useful
- 2.2Designing tool schemas — names, descriptions, parameters
- 2.3Tool execution: synchronous, async, parallel, conditional
- 2.4Building common tools: web search, database, file ops, API calls
- 2.5Tool error handling: retries, fallbacks, graceful degradation
Module 3: Memory Systems
- 3.1Why agents need memory (beyond the context window)
- 3.2Short-term memory: conversation history and sliding windows
- 3.3Long-term memory: vector databases, knowledge graphs
- 3.4Episodic memory: remembering past experiences and outcomes
- 3.5Memory retrieval: similarity search, recency bias, importance scoring
Module 4: Planning & Reasoning
- 4.1Planning strategies: top-down decomposition, iterative refinement
- 4.2ReAct pattern implementation
- 4.3Plan-and-execute: separate planning from execution
- 4.4Reflection and self-correction
- 4.5Human-in-the-loop: when to ask for help
Module 5: Building Agents with LangGraph
- 5.1LangGraph architecture — states, nodes, edges, conditional routing
- 5.2Your first LangGraph agent
- 5.3State management — passing data through the graph
- 5.4Conditional edges — dynamic routing
- 5.5Checkpointing and persistence
- 5.6LangGraph Studio — visual debugging
Module 6: Building Agents with CrewAI
- 6.1CrewAI philosophy — roles, tasks, and collaborative agents
- 6.2Defining agents with roles, goals, and backstories
- 6.3Task definition and delegation patterns
- 6.4Crew orchestration — sequential, hierarchical, consensual
- 6.5Custom tools and external service integration
Module 7: Building Agents with AutoGen & OpenAI Agents SDK
- 7.1AutoGen's conversation-based approach
- 7.2GroupChat patterns — multiple agents discussing
- 7.3OpenAI Agents SDK — the simplest path for OpenAI-native apps
- 7.4Swarm pattern — lightweight handoff-based coordination
- 7.5Framework comparison matrix — when to use which
Module 8: Claude Agent SDK & Anthropic Ecosystem
- 8.1Claude Agent SDK (TypeScript & Python)
- 8.2Building agents with extended thinking
- 8.3MCP integration — connecting agents to any data source
- 8.4Skills — packaging reusable agent capabilities
- 8.5Subagent delegation — main agent to specialized subagents
Module 9: Project 1 — Personal Research Agent
- 9.1Architecture: research query, web search, document analysis, report
- 9.2Building the search and retrieval layer
- 9.3Source evaluation and credibility scoring
- 9.4Report generation with citations
- 9.5Iterative research: identifying gaps and searching again
Module 10: Project 2 — Customer Service Agent System
- 10.1Multi-agent architecture: router, FAQ, ticket, escalation agents
- 10.2Knowledge base integration
- 10.3Sentiment detection and dynamic routing
- 10.4Escalation logic: when to involve a human
- 10.5Conversation quality monitoring and feedback loops
Module 11: Project 3 — Code Generation & Review Agent
- 11.1Building a coding agent that reads, writes, and executes code
- 11.2Sandboxed code execution
- 11.3Automated testing — agent writes code AND tests
- 11.4Code review agent — analyzing PRs for bugs and security
- 11.5Integration with GitHub/GitLab APIs
Module 12: Production Deployment & Operations
- 12.1Deploying agents — Docker, cloud functions, dedicated servers
- 12.2Cost management — token budgets, caching, model selection
- 12.3Observability — logging, tracing, monitoring
- 12.4Error recovery — graceful failures, retries, circuit breakers
- 12.5Scaling — concurrent agent runs and queue management
Module 13: Safety, Guardrails & Evaluation
- 13.1Why agent safety matters — real examples of agents going wrong
- 13.2Input validation and prompt injection defense
- 13.3Output guardrails — preventing harmful or incorrect outputs
- 13.4Sandboxing — limiting what agents can access
- 13.5Evaluation frameworks — measuring agent performance
Module 14: Project 4 — Multi-Agent Business Automation
- 14.1Architecture: orchestrator, data, analysis, report, notification agents
- 14.2Inter-agent communication protocols
- 14.3Shared state and coordination patterns
- 14.4Failure handling in multi-agent systems
- 14.5End-to-end testing of multi-agent workflows
Module 15: Your Custom Agent (Capstone)
- 15.1Choosing your agent project (guided brainstorming)
- 15.2Architecture design review
- 15.3Implementation sprint with checkpoints
- 15.4Testing, deployment, and documentation
- 15.5Portfolio showcase — presenting your agent to the community
AI isn't slowing down.
Neither should you.
Every week you wait, the gap widens. The people who invest in learning AI now will be the ones leading teams, building companies, and staying ahead of the curve. This is your moment — don't let it pass.