Testing, Limits, and Next Steps

Testing Your Agent

Try these test cases:

Test Case	Expected Behavior
”Add two tasks then list”	Should add both, then list them
”Mark a non-existing ID complete”	Should handle gracefully
”What tasks do I have?”	Should call listTodos
”I need to remember dentist on Tuesday”	Should extract date and add

Current Limits

• In-memory store only - Resets on restart
• No authentication - Anyone can use it
• Simple tools only - Just the three basic operations
• No error recovery - Fails if API is down

Suggested Extensions

Here are ideas to make it better:

1. Persist to database - Use SQLite or PostgreSQL
2. Add more tools - remindLater, updateTodo, searchTodos
3. Add fields - Priority, tags, notes
4. Streaming responses - Show responses as they generate
5. Better error handling - Retry logic, fallbacks

Challenge

Try adding a new tool: clearCompleted().

1. Add the function to tools.ts
2. Add the tool schema to agent.ts
3. Handle it in the tool execution loop
4. Test it with: “Clear all completed tasks”

---

Summary

What You Built

You built a working AI agent with:

• ✅ Tools - Three functions for managing todos
• ✅ LLM Integration - Function calling with OpenAI
• ✅ Agent Loop - The core loop that calls tools and responds
• ✅ Memory - In-memory state that persists across turns
• ✅ Basic UI - A simple web interface

Key Takeaways

1. Agents are loops - They process messages, call tools, and loop until done
2. Tools are functions - Simple JavaScript functions that the LLM can call
3. Function calling - The LLM decides which tools to use based on schemas
4. Memory matters - State needs to persist across turns

The Mental Model

Agent = Model + Tools + Memory + Loop

• Model: The LLM that makes decisions
• Tools: Functions the agent can call
• Memory: State that persists
• Loop: The process that ties it all together

What’s Next?

• Build more complex agents with multiple tools
• Add persistent storage
• Create production-ready UIs
• Explore agent frameworks (LangChain, AutoGPT, etc.)

You’ve got the fundamentals. Now go build something cool!

---

Final Knowledge Check

Let’s make sure you understand everything:

---

---

Congratulations! You’ve built your first AI agent. You now understand how agents work, how to define tools, and how to wire them to an LLM.

Keep experimenting, and happy building! 🚀

---