A chatbot talks. An agent acts. Learn how LLMs use tools, plan steps, and interact with the real world.
Agents add three superpowers to standard LLMs: Planning, Memory, and Tools.
(Click to reveal)
Agents can call external functions (APIs, Calculators, Google Search) to do things the LLM can't do alone.
(Click to reveal)
Breaking a big goal ("Book a vacation") into smaller, manageable steps ("Search flights", "Book hotel").
(Click to reveal)
Short-term memory (conversation history) and Long-term memory (Vector DBs) allow agents to remember past actions.
(Click to reveal)
A prompting pattern where the model "Thinks" about what to do, "Acts" (calls a tool), and "Observes" the output.
Step through the lifecycle of an autonomous agent solving a problem.
Click on the nodes in the circle (Thought -> Action -> Observation) to simulate the agent's reasoning.
The LLM analyzes the goal. It realizes it doesn't know the current weather in Tokyo because its training data is old.
The agent constructs a formatted command (often JSON) to call an external tool.
The tool executes and returns real-world data. The agent reads this output.
(The loop creates a new Thought: "I now have the answer.")
Agents use advanced prompting strategies to break down complex tasks.
Thinking step-by-step to avoid logical errors.
Exploring multiple possibilities (branches) and selecting the most promising one.
Reviewing past actions to self-correct ("Wait, that search failed, let me try a different keyword").
Test your Agentic knowledge.
1. What is the key difference between a Chatbot and an Agent?
2. What does "ReAct" stand for?
3. Why does an agent need "Memory"?