Loading...
Loading...
From simple chat to autonomous agents — perception, reasoning, action cycles and the agent loop
| Aspect | Chatbot | Agent |
|---|---|---|
| Interaction | Respond once | Multi-step loop |
| Tools | None | Can call APIs, run code |
| Memory | Current conversation | Persistent memory |
| Autonomy | None | Can plan independently |
1. PERCEIVE: Receive request + current state
2. THINK: Reason about what to do next
3. ACT: Call a tool or produce output
4. OBSERVE: See the result
5. REPEAT: Until goal achieved| Level | Description |
|---|---|
| L0 | Direct LLM call, single response |
| L1 | Can use tools, user approves |
| L2 | Semi-autonomous, asks confirmation |
| L3 | Full autonomy within scope |
Thought: I need to find the customer's order.
Action: search_database(order='12345')
Observation: Order found — shipped.
Thought: The customer wants the tracking number.
Action: get_tracking(order='12345')
Observation: Tracking: 1Z999...
Final: Your order has shipped! Tracking: 1Z999...Design an agent for trip planning. What steps and tools would it need?