Agentic Challenges

Try these after completing all 5 agentic steps.

Knowledge Check

  1. What is the difference between a regular LLM call and an agentic LLM call?

  2. What does tool_choice="auto" mean in the OpenAI API?

  3. Why do we limit the agent loop to a maximum number of turns?

  4. What is the purpose of tool_call_id in the tool response message?

  5. How does chain-of-thought prompting improve AI reasoning?

Code Challenges

Challenge 1: Write Tests First (TDD)

Practice test-driven development: write the test for the weather tool before you build the tool.

def test_get_weather_known_city():
    from agent import get_weather
    result = get_weather("Paris")
    assert "°C" in result


def test_get_weather_unknown_city():
    from agent import get_weather
    result = get_weather("Atlantis")
    assert "not available" in result

Run the tests — they will fail because get_weather doesn't exist yet. Now build the tool and make them pass.

Challenge 2: Weather Agent

Add a real weather tool that calls a free weather API (like wttr.in or OpenWeatherMap) instead of the mock get_weather function.

Hint: wttr.in doesn't need an API key:

response = requests.get(f"https://wttr.in/{city}?format=%C+%t")

Challenge 2: SQL Query Agent

Build a tool that lets the AI query the SQLite database directly:

def query_database(sql: str) -> str:
    """Execute a SQL query and return results."""
    # Only allow SELECT statements for safety
    ...

Then test: "How many messages have I sent?" or "Show me all messages from today."

Challenge 3: Multi-Memory Agent

Add a "remember" and "forget" tool so the AI can store and recall facts across conversations:

User: "Remember that my name is Alice"
AI: "I'll remember that."  (calls remember("name", "Alice"))

User: "What's my name?"
AI: "Your name is Alice."  (calls recall("name") → "Alice")

Store memories in a dictionary or a simple JSON file.

Open-Ended Project: Custom Research Assistant

Build a research assistant that can:

  1. Search Wikipedia for background information
  2. Search a custom knowledge base (a JSON file with facts)
  3. Summarize findings into a report
  4. Save the report to a file

Requirements: - Agent takes a topic as input - Searches multiple sources - Compiles a structured report with sections - Saves the report to a .md file

Stretch goals: - Add a "generate image" tool using DALL-E - Add a "translate" tool using a translation API - Add a "send email" tool (mock) - Build a web UI for the research agent using Flask

Example flow:

User: "Research renewable energy and write a report"

1. search_wikipedia("Solar power") → summary
2. search_wikipedia("Wind power") → summary
3. search_wikipedia("Hydroelectric power") → summary
4. save_report("renewable-energy.md", report_text)
5. AI: "I've written a report on renewable energy to renewable-energy.md"
← Chat Enhancement