Core Concepts

LLMs alone can only produce text. Real work requires calling APIs, querying databases, browsing the web, writing files, and coordinating multi-step workflows. Agents wrap LLMs in an execution loop with tools and memory, turning a chatbot into a system that actually gets things done.

A chatbot is a consultant who gives advice from a chair. An agent is a consultant who gets up, logs into your systems, pulls the data, drafts the report, sends it for review, and follows up — reporting back at each step.

Scenario

User: 'Analyze our Q3 sales data and email the summary to the leadership team.' Agent: queries SQL database → aggregates results → generates chart → drafts email → requests human approval → sends via email API → confirms completion.

Solution

In Agent Foundations, apply Core Concepts to this scenario: User: 'Analyze our Q3 sales data and email the summary to the leadership team. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Core Concepts (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where Core Concepts happens in the code.

1from openai import OpenAI  # import dependencies2
3client = OpenAI()  # create API client4
5tools = [6    {"type": "function", "function": {7        "name": "query_database",8        "description": "Run a SQL query on the sales database",9        "parameters": {"type": "object", "properties": {10            "sql": {"type": "string"}11        }, "required": ["sql"]}12    }},13    {"type": "function", "function": {14        "name": "send_email",15        "description": "Send an email to recipients",16        "parameters": {"type": "object", "properties": {17            "to": {"type": "string"}, "subject": {"type": "string"}, "body": {"type": "string"}18        }, "required": ["to", "subject", "body"]}19    }},20]21
22messages = [{"role": "user", "content": "Analyze Q3 sales and email leadership"}]23# Agent loop: call LLM → if tool_calls → execute → append result → repeat

• No iteration limit — agent runs forever on impossible tasks
• Giving agents unrestricted tool access without sandboxing
• No observability — impossible to debug agent failures
• Treating agents as chatbots with one extra API call
• No human approval for destructive actions (delete, send, pay)

Quick recap — the most important points from this module.