The Build an Agent tutorial needs additional work

[dom-ar]

The agent in this tutorial needs a prompt for sure and the tool needs to be described.

For example with the existing code requests like What is the weather in ____ did not work.
However creating the tavily as a custom tool (I actually don't know if Tavily already has a description) and adding a prompt makes it work way better.

[onestardao]

Totally agree — that tutorial feels like it skipped the part where the agent gets a brain 😅
Without a prompt or a description, it’s like sending an intern out to do weather forecasts with no training.

I ran into the same “What is the weather in ___” issue. Turns out the tool didn’t even know what it was supposed to be.
Giving it a proper prompt helped a ton. Something like:

Tool(
  name="weather_tool",
  description="Use this to answer weather-related questions. Example: 'What's the weather in Paris?'",
  func=...
)

This actually makes it usable in practice.
LangChain’s agent framework is powerful but assumes you know what not to omit... which isn’t fair to people learning by example.

Thanks for bringing this up — more people should bump into this thread before rage-uninstalling