Complete AI Mastery Course

{"contentFormat":"slides.v1","completion":{"requireAllSlides":true,"requireQuiz":true},"slides":[{"kind":"title","eyebrow":"Module 7: Agentic AI and Autonomous Systems","title":"Building Your First AI Agent: Tools and Frameworks","body":"Move from theory to practice. Learn the anatomy of an agent, choose the right framework, and build a simple, safe agent with free tools. No expert programming required.","outcomes":["Identify the four core components of a buildable agent","Compare frameworks: LangChain, LangGraph, CrewAI, AutoGen, and provider-native tool use","Write effective tool descriptions that guide model behavior","Set goals, stop conditions, and cost controls for reliable operation","Build a minimal single-tool agent step by step"],"narration":"Welcome to Module 7. In this lesson, you'll learn how to build your first AI agent. We'll cover the essential components, compare popular frameworks, and walk through a hands-on example. Let's get started."},{"kind":"content","heading":"The Anatomy of a Buildable Agent","body":"Every agent combines four components plus a loop:\n\n- Model: A capable LLM (e.g., GPT-4o, Claude 3.5) accessed via API or run locally.\n- Tools: Functions the model can call. Each tool needs a name, description, and defined inputs. Modern APIs use function calling.\n- Memory: At minimum, the conversation history; optionally a vector store for long-term recall.\n- Loop: Code that feeds context, executes tool calls, returns results, and repeats until a stop condition.\n\nAnalogy: Setting up a new employee's desk – skills (model), tools (functions), notebook (memory), shift routine (loop with stop condition).","callout":{"variant":"insight","title":"Key Insight","text":"The model decides which tool to call purely from your description. Writing clear, specific tool descriptions is the highest-leverage part of building an agent."},"narration":"An agent has four parts: a model, tools, memory, and a loop. The model uses tool descriptions to decide when to call a function. Good descriptions are critical."},{"kind":"content","heading":"Choosing a Framework (as of 2026)","body":"You can write the loop yourself, but frameworks save time. Here are the main options:\n\n| Framework | Best For | Notes |\n|-----------|----------|-------|\n| LangChain | Broad toolkit, flexible | Well-documented; can be heavy for small projects |\n| LangGraph | Explicit graph control, reliability | Fine-grained loops, human-in-the-loop |\n| CrewAI | Multi-agent teams | Role-based, fast to prototype |\n| AutoGen | Conversation-driven multi-agent | Strong for agent-agent and human-agent messaging |\n| Provider-native (OpenAI, Anthropic) | Learning mechanics, minimal dependencies | Use SDK tool calling directly |\n\nGuidance: For your first agent, start with provider-native tool calling or a lightweight framework. Avoid large multi-agent systems initially.","callout":{"variant":"tip","title":"Recommendation","text":"Start with provider-native tool calling to understand the mechanics. Then move to LangGraph if you need explicit control or CrewAI for multi-agent scenarios."},"narration":"You can build an agent with just an SDK, or use frameworks like LangChain, LangGraph, CrewAI, or AutoGen. For your first agent, keep it simple."},{"kind":"content","heading":"Writing Effective Tool Descriptions","body":"The model relies on your description to decide when to call a tool. Follow these principles:\n\n- Be specific: \"Look up the current exchange rate between two three-letter currency codes\" is better than \"currency tool.\"\n- State inputs and outputs clearly: Include parameter types and expected return format.\n- Say when not to use it: E.g., \"Do not use for historical data.\"\n\nExample (OpenAI function calling):\n\n

json\n{\n  \"name\": \"get_weather\",\n  \"description\": \"Get current temperature and conditions for a given city. Use when user asks about weather.\",\n  \"parameters\": {\n    \"type\": \"object\",\n    \"properties\": {\n      \"city\": {\n        \"type\": \"string\",\n        \"description\": \"City name, e.g., Nairobi\"\n      }\n    },\n    \"required\": [\"city\"]\n  }\n}\n

\n\nMinimal code snippet (Python with OpenAI):\n\n

python\nimport openai\n\ntools = [\n    {\n        \"type\": \"function\",\n        \"function\": {\n            \"name\": \"get_weather\",\n            \"description\": \"Get current temperature and conditions for a given city.\",\n            \"parameters\": {\n                \"type\": \"object\",\n                \"properties\": {\n                    \"city\": {\"type\": \"string\", \"description\": \"City name\"}\n                },\n                \"required\": [\"city\"]\n            }\n        }\n    }\n]\n\nmessages = [{\"role\": \"user\", \"content\": \"What's the weather in Jakarta?\"}]\nresponse = openai.chat.completions.create(\n    model=\"gpt-4o\",\n    messages=messages,\n    tools=tools\n)\n# Check if tool call is requested\nif response.choices[0].message.tool_calls:\n    # Execute function and append result\n    pass\n

","callout":{"variant":"exercise","title":"Try It Yourself","text":"Write a tool description for a function that looks up product stock by SKU. Be specific about inputs and when to use it."},"narration":"Tool descriptions are crucial. Be specific, state inputs and outputs, and mention when not to use the tool. The code example shows how to define a tool with OpenAI."},{"kind":"content","heading":"Reliability, Cost, and Observability","body":"Building a reliable agent requires planning for failure and controlling costs.\n\n- Observability first: Log every thought, tool call, and result. You cannot debug an agent you cannot see.\n- Cost control: Each loop step is a paid model call. Set a step cap (e.g., max 5 steps) and, if possible, a spending cap via API limits.\n- Stop conditions: Define a concrete goal (e.g., \"user question answered\") and a hard fallback (max steps).\n- Graceful failure: If the agent cannot complete the task, it should say so clearly rather than invent an answer.\n\nReal-World Examples:\n\n- Small shop in Nigeria: A single-tool agent for product lookup on WhatsApp. One tool, one goal, fully observable.\n- Student in Indonesia: A study agent with web search and note-saver, capped at 8 steps to keep costs near zero.\n- Consultant in UK: Uses LangGraph with a human-in-the-loop node for approval before sending emails.","callout":{"variant":"warning","title":"Common Pitfall","text":"No step limit can lead to runaway loops and high costs. Always set a maximum step count and a clear success condition."},"narration":"Observability, cost control, and clear stop conditions are essential. Learn from real-world examples: start small, cap steps, and log everything."},{"kind":"quiz","heading":"Check Your Understanding","questions":[{"question":"Which component of an agent is responsible for deciding when to call a tool?","options":["Memory","The loop","The model","The tool itself"],"questionId":"cmrf73kb8002mpd27yrrri5ki"},{"question":"What is the recommended first step for building your first agent?","options":["Use a multi-agent framework like CrewAI","Start with provider-native tool calling or a lightweight framework","Build a complex system with five agents and ten tools","Skip tool descriptions to save time"],"questionId":"cmrf73kb8002npd27zhjbh2v7"},{"question":"How can you prevent an agent from looping indefinitely and burning money?","options":["Use a more expensive model","Set a maximum step count and a clear success condition","Disable logging","Remove the stop condition"],"questionId":"cmrf73kb8002opd27gxxuqgfy"}],"narration":"Let's test your understanding with a quick quiz. Answer the questions to reinforce the key concepts.","quizId":"qz_cmk7lm8oh002zg4p89hpln47g"},{"kind":"summary","heading":"Key Takeaways","takeaways":["An agent consists of a model, tools, memory, and a loop.","Tool descriptions are critical: be specific, state inputs/outputs, and mention when not to use.","Start with provider-native tool calling or a lightweight framework for your first agent.","Always set a step cap and success condition to control costs and prevent runaway loops.","Observability (logging) is essential for debugging and reliability.","Build for graceful failure: a clear 'I couldn't complete this' is better than a confident wrong answer."],"narration":"To summarize: know the four components, write good tool descriptions, start simple, control costs, and log everything. Build small, iterate, and grow from success."}]}