How do agents act in the outside world?
Why D is the answer
Through tools such as extensions, functions and data stores
Agents act through tools, which include extensions, functions and data stores.
In detail
- Agents act through tools such as extensions, functions and data stores.
- Without tools, an agent can only produce text.
Use case
An illustrative scenario, not a real incident.
An agent checks stock in a database, and it calls a shipping function, through its tools.
See it in different ways
sequenceDiagram participant U as User participant A as Agent participant T as Tool or data store U->>A: Goal A->>T: Call a function or extension T-->>A: Result A-->>U: Answer or action
sequenceDiagram participant U as User participant A as Agent participant T as Tool or data store U->>A: Goal A->>A: Reasons about the next step A->>T: Uses a tool T-->>A: Result A-->>U: Outcome
mindmap
root(("Agent tools"))
n1["Types"]
n2["Extensions"]
n3["Functions"]
n4["Data stores"]Key point: Tools let agents act.
Why the other options are wrong
- A. By training a new model for every request
- Training per request is not how agents work.
- B. By rewriting their own system prompt
- Rewriting a prompt does not act on anything.
- C. By asking users to do each step by hand
- That defeats the purpose.