Agents Best Practices

There are many aspects to take into consideration when developing AI agent. In this page, I try to summarize tips and tricks from the sources specified above. Additionally, I also include some of my personal insights in developing these systems.

Each use of a LLM introduces uncertainty and risk into the system. Therefore, LLM usage should be kept at a minimum. Furthermore, go through this checklist when deciding whether to employ LLM for a given component of the system:

1. Can it be implemented with software engineering?
2. Can it be achieved with traditional machine learning?
3. If answer to both questions above is no, then use a LLM.

Only employ a framework when you understand underlying abstraction that it implements. Otherwise, the solution can become hard to debug and modify. Ideally, try to implement as many parts of the solution as possible without framework ( Anthropic - Building effective agents).

Building good agents:

Make your task very clear! Since an agent is powered by an LLM, minor variations in your task formulation might yield completely different results.
Then, improve the information flow towards your agent in tool use.

Particular guidelines to follow:

• Each tool should log (by simply using print statements inside the tool’s forward method) everything that could be useful for the LLM engine.
  • In particular, logging detail on tool execution errors would help a lot!