{"data":{"_id":"6ae3f434ebf197f2924ed34f","name":"Agentic AI","slug":"agentic-ai","image":"https://images.unsplash.com/photo-1677442136019-21780ecad995?auto=format&fit=crop&w=1200&q=80","icon":"ai-ml","accentColor":"#8B5CF6","isFree":false,"price":1499,"features":["Model Context Protocol (MCP) Standards","LangGraph Cyclic State Architectures","Production Hybrid RAG & Cross-Encoders","Autonomous Self-Healing Coding Agents","Docker & Wasm Execution Sandboxes"],"tags":["Interactive Roadmap","Agentic AI","MCP","LLMs","LangGraph","Autonomous Systems"],"category":{"_id":"6ae3f433ebf197f2924ed34e","name":"AI & Machine Learning","description":"Cutting-edge Generative AI, LLMs, and Agentic Systems"},"description":"Master Agentic AI from first principles to advanced production architectures. Build autonomous agents, MCP tool servers, LangGraph state machines, and self-healing coding swarms.","modules":[{"icon":"🧠","title":"Module 1: Foundations of Agentic AI (From Scratch)","description":"Understand the shift from passive LLM prompt-response chats to autonomous goal-driven cognitive agents.","duration":"2 hours","order":1,"isFree":true,"isPublished":true,"isProject":false,"content":"### Welcome to Agentic AI\nIn this foundational module, you transition from simple prompt engineering to architecting systems that reason, make decisions, execute tools, and pursue multi-step goals autonomously.\n\n```\n[User Goal] ──> [Agent Brain (LLM)] ──> [Tool Execution] ──> [Observation] ──> [Autonomous Completion]\n```","subsections":[{"id":"sub-1-1","title":"1. What Makes an AI \"Agentic\"?","description":"The fundamental difference between passive chatbots and proactive, autonomous agents.","duration":"20 mins","content":"## What Makes an AI \"Agentic\"?\n\n### The Paradigm Shift\nTraditional generative AI is **passive**: you send a prompt, it returns a text completion, and the process terminates. An **Agentic AI** system is **autonomous and goal-driven**: given an objective, it plans a sequence of actions, perceives feedback from its environment, and iterates until the goal is achieved.\n\n```\nTraditional LLM:\n[User Prompt] ────────────> [LLM] ────────────> [Static Answer]\n\nAgentic System:\n                ┌───────── Perception ──────────┐\n                ▼                               │\n[Goal] ──> [Agent Core] ──> [Plan] ──> [Tool Action] ──> [Environment]\n                │                               ▲\n                └─────── Evaluation Loop ───────┘\n```\n\n### The 4 Pillars of an Agent\n1. **Perception**: Reading environment state, user inputs, API responses, or code execution errors.\n2. **Reasoning**: Deciding what to do next based on goals and past observations.\n3. **Action (Tool Use)**: Modifying files, issuing database queries, or hitting third-party APIs.\n4. **Learning & Memory**: Storing conversation history and episodic learnings to avoid repeating mistakes.\n\n> 💡 **Key Takeaway**: A chatbot answers questions. An agent completes tasks.","isFree":true,"isPublished":true,"order":1,"_id":"6ae3f434ebf197f2924ed351","locked":false},{"id":"sub-1-2","title":"2. The Core Anatomy of an AI Agent","description":"Deconstructing the internal components: Brain, Sensors, Actuators, and Memory.","duration":"25 mins","content":"## The Core Anatomy of an AI Agent\n\n### How the Pieces Fit Together\nEvery modern AI agent—from Devin to AutoGPT—is built out of four foundational subsystems:\n\n```\n┌────────────────────────────────────────────────────────┐\n│                   AGENT COGNITIVE CORE                 │\n├─────────────────┬──────────────────────────────────────┤\n│ 1. Brain (LLM)  │ GPT-4o, Claude 3.5 Sonnet, Gemini 1.5│\n├─────────────────┼──────────────────────────────────────┤\n│ 2. Memory       │ Short-term (Context) & Long-term (DB)│\n├─────────────────┼──────────────────────────────────────┤\n│ 3. Planning     │ Decomposing goals into sub-tasks     │\n├─────────────────┼──────────────────────────────────────┤\n│ 4. Tools/MCP    │ Bash, Git, PostgreSQL, REST APIs     │\n└─────────────────┴──────────────────────────────────────┘\n```\n\n### Component Breakdown\n| Component | Function | Real-World Example |\n| :--- | :--- | :--- |\n| **Brain** | Natural language reasoning & decision making | Claude 3.5 Sonnet / GPT-4o |\n| **Short-Term Memory** | Current session context & tool execution traces | Message thread history |\n| **Long-Term Memory** | Vector database & factual memory storage | Pinecone, ChromaDB, pgvector |\n| **Actuators (Tools)** | Concrete interfaces that change external state | File Writer, Terminal Runner, HTTP Client |\n\n> ⚡ **Mental Model**: Think of the LLM as the *CPU*, Memory as *RAM/Disk*, and Tools as *Peripherals* (Keyboard, Screen, Network).","isFree":true,"isPublished":true,"order":2,"_id":"6ae3f434ebf197f2924ed352","locked":false},{"id":"sub-1-3","title":"3. The ReAct Pattern: Reasoning + Acting","description":"Implementing the industry standard Thought-Action-Observation loop.","duration":"35 mins","content":"## The ReAct Pattern: Reasoning + Acting\n\n### How ReAct Works\nIntroduced by Yao et al., **ReAct** (Reasoning + Acting) interleaves verbal reasoning (\"Thought\") with concrete execution (\"Action\") and feedback (\"Observation\").\n\n```\n1. Thought:     \"I need to find the user's latest invoice in Stripe.\"\n2. Action:      stripe.getInvoice({ customerId: \"cus_123\" })\n3. Observation: { id: \"in_456\", amount: 4900, status: \"paid\" }\n4. Thought:     \"The invoice is paid. I will now format the email.\"\n5. Action:      sendEmail({ to: \"user@example.com\", subject: \"Invoice paid\" })\n6. Final Ans:   \"Invoice #in_456 has been confirmed and sent.\"\n```\n\n### Minimal ReAct Loop in TypeScript\n\n```typescript\ninterface Tool {\n  name: string;\n  execute: (input: string) => Promise<string>;\n}\n\nasync function runReActAgent(goal: string, tools: Record<string, Tool>) {\n  let context = `Goal: ${goal}\\n`;\n  \n  for (let step = 0; step < 5; step++) {\n    // 1. LLM generates Thought and next Action\n    const response = await callLLM(context + \"Thought: What is next step?\\nAction: toolName(arg)\");\n    const { thought, toolName, arg } = parseAction(response);\n    \n    if (toolName === \"FINISH\") return arg;\n\n    // 2. Execute tool\n    const observation = await tools[toolName].execute(arg);\n    \n    // 3. Feed back into loop\n    context += `Thought: ${thought}\\nAction: ${toolName}(${arg})\\nObservation: ${observation}\\n`;\n  }\n}\n```\n\n> 💡 **Key Takeaway**: Without \"Thought\", the agent takes blind actions. Without \"Observation\", it hallucinates outcomes. The loop ties reasoning to real-world truth.","isFree":true,"isPublished":true,"order":3,"_id":"6ae3f434ebf197f2924ed353","locked":false}],"_id":"6ae3f434ebf197f2924ed350","locked":false},{"icon":"🔧","title":"Module 2: Structured Tool Use & Function Calling","description":"Master JSON Schema validation, Pydantic tool contracts, and Anthropic’s Model Context Protocol (MCP)....","duration":"2.5 hours","order":2,"isFree":false,"isPublished":true,"isProject":false,"content":"","subsections":[{"id":"sub-2-1","title":"1. Strict Function Calling with Schemas","description":"Enforce deterministic JSON outputs and eliminate parameter hallucination.","duration":"30 mins","content":"","isFree":false,"isPublished":true,"order":1,"_id":"6ae3f434ebf197f2924ed355","locked":true},{"id":"sub-2-2","title":"2. Model Context Protocol (MCP) Architecture","description":"The open standard for connecting AI agents to data sources, files, and developer tools.","duration":"35 mins","content":"","isFree":false,"isPublished":true,"order":2,"_id":"6ae3f434ebf197f2924ed356","locked":true},{"id":"sub-2-3","title":"3. Building a Production MCP Tool Server","description":"Step-by-step implementation of an MCP server with stdio transport.","duration":"40 mins","content":"","isFree":false,"isPublished":true,"order":3,"_id":"6ae3f434ebf197f2924ed357","locked":true}],"_id":"6ae3f434ebf197f2924ed354","locked":true},{"icon":"💾","title":"Module 3: Memory Systems & Context Engineering","description":"Design multi-tiered memory architectures: Working memory, Episodic recall, and Hybrid RAG with Cross-Encoders....","duration":"2.5 hours","order":3,"isFree":false,"isPublished":true,"isProject":false,"content":"","subsections":[{"id":"sub-3-1","title":"1. Working vs Episodic vs Semantic Memory","description":"The three tiers of agent cognition modeled after human cognitive psychology.","duration":"25 mins","content":"","isFree":false,"isPublished":true,"order":1,"_id":"6ae3f434ebf197f2924ed359","locked":true},{"id":"sub-3-2","title":"2. Hybrid RAG: Vectors + BM25 + Reranking","description":"Why naive vector search fails in production agents and how to build a 3-stage retrieval pipeline.","duration":"35 mins","content":"","isFree":false,"isPublished":true,"order":2,"_id":"6ae3f434ebf197f2924ed35a","locked":true},{"id":"sub-3-3","title":"3. Context Compression & Token Pruning","description":"Keep agent context concise across hundreds of loop iterations.","duration":"30 mins","content":"","isFree":false,"isPublished":true,"order":3,"_id":"6ae3f434ebf197f2924ed35b","locked":true}],"_id":"6ae3f434ebf197f2924ed358","locked":true},{"icon":"🧭","title":"Module 4: Planning, Reflection & Self-Correction","description":"Implement Plan-and-Solve architectures, Reflexion self-critique loops, and Human-in-the-Loop approval checkpoints....","duration":"3 hours","order":4,"isFree":false,"isPublished":true,"isProject":false,"content":"","subsections":[{"id":"sub-4-1","title":"1. Plan-and-Solve vs Tree-of-Thought","description":"Architecting agents that formulate, validate, and dynamically update multi-step plans.","duration":"30 mins","content":"","isFree":false,"isPublished":true,"order":1,"_id":"6ae3f434ebf197f2924ed35d","locked":true},{"id":"sub-4-2","title":"2. The Reflexion Loop: Automated Self-Correction","description":"Teaching agents to analyze their own failures and fix buggy code autonomously.","duration":"35 mins","content":"","isFree":false,"isPublished":true,"order":2,"_id":"6ae3f434ebf197f2924ed35e","locked":true},{"id":"sub-4-3","title":"3. Human-in-the-Loop (HITL) & Safety Guardrails","description":"Building pause-and-resume approval gates for sensitive operations.","duration":"30 mins","content":"","isFree":false,"isPublished":true,"order":3,"_id":"6ae3f434ebf197f2924ed35f","locked":true}],"_id":"6ae3f434ebf197f2924ed35c","locked":true},{"icon":"🐝","title":"Module 5: Multi-Agent Orchestration & Swarms","description":"Move from single agents to multi-agent teams using LangGraph, specialized worker swarms, and consensus protocols....","duration":"3.5 hours","order":5,"isFree":false,"isPublished":true,"isProject":false,"content":"","subsections":[{"id":"sub-5-1","title":"1. Multi-Agent Topologies & Patterns","description":"Supervisors, peer-to-peer swarms, and sequential handoff architectures.","duration":"35 mins","content":"","isFree":false,"isPublished":true,"order":1,"_id":"6ae3f434ebf197f2924ed361","locked":true},{"id":"sub-5-2","title":"2. Cyclic State Graphs with LangGraph","description":"Building production state machines with typed channels and conditional edges.","duration":"40 mins","content":"","isFree":false,"isPublished":true,"order":2,"_id":"6ae3f434ebf197f2924ed362","locked":true},{"id":"sub-5-3","title":"3. Inter-Agent Communication Protocols","description":"Structured message passing, state handoffs, and consensus mechanisms.","duration":"35 mins","content":"","isFree":false,"isPublished":true,"order":3,"_id":"6ae3f434ebf197f2924ed363","locked":true}],"_id":"6ae3f434ebf197f2924ed360","locked":true},{"icon":"🛡️","title":"Module 6: Observability, Evaluation & Sandboxing","description":"Productionize agents with OpenTelemetry distributed tracing, SWE-bench evaluations, and secure Docker sandboxes....","duration":"2.5 hours","order":6,"isFree":false,"isPublished":true,"isProject":false,"content":"","subsections":[{"id":"sub-6-1","title":"1. Distributed Tracing & Cost Observability","description":"Instrument agent loops with OpenTelemetry, Langfuse, and step-level latency tracking.","duration":"30 mins","content":"","isFree":false,"isPublished":true,"order":1,"_id":"6ae3f434ebf197f2924ed365","locked":true},{"id":"sub-6-2","title":"2. Secure Code Sandboxing (Docker & Wasm)","description":"Isolate agent code execution to prevent system compromise and malicious commands.","duration":"35 mins","content":"","isFree":false,"isPublished":true,"order":2,"_id":"6ae3f434ebf197f2924ed366","locked":true}],"_id":"6ae3f434ebf197f2924ed364","locked":true},{"icon":"🚀","title":"Module 7: Capstone: Autonomous Full-Stack AI Engineer","description":"Build a production-grade autonomous coding agent that analyzes repositories, reproduces bugs, and submits verified Pull Requests....","duration":"4 hours","order":7,"isFree":false,"isPublished":true,"isProject":false,"content":"","subsections":[{"id":"sub-7-1","title":"1. Architecture of an Autonomous Dev Agent","description":"End-to-end system design: Repo indexer, semantic symbol graph, and file patch engine.","duration":"45 mins","content":"","isFree":false,"isPublished":true,"order":1,"_id":"6ae3f434ebf197f2924ed368","locked":true},{"id":"sub-7-2","title":"2. Self-Healing Build Loops & Pull Request Delivery","description":"Iterative lint fixing, type checking, test validation, and automated GitHub PR creation.","duration":"45 mins","content":"","isFree":false,"isPublished":true,"order":2,"_id":"6ae3f434ebf197f2924ed369","locked":true}],"_id":"6ae3f434ebf197f2924ed367","locked":true}],"details":[{"label":"Overview","content":[{"title":"From Chatbots to Autonomous Agents","description":"A comprehensive, hands-on path taking you from foundational ReAct loops to enterprise-grade multi-agent swarms and secure sandbox execution.","_id":"6ae3f434ebf197f2924ed36b","locked":false}],"_id":"6ae3f434ebf197f2924ed36a"}],"status":"published","__v":0},"message":"Course fetched successfully"}