Overview

The following section provides an overview of various common prompt techniques. These can significantly influence the output of LLMs and agents in CompanyGPT and lead to the desired result.

The prompt techniques presented here can all be implemented very well in CompanyGPT. It is very important to note that these prompt techniques are not either/or, but rather interdependent. Powerful AI applications and agents always use a combination (often) of all the techniques presented here.

Example agent prompt (structural):

• Prompt reasonably structured
• Prompt formatted to emphasize certain aspects of the instructions
• Few-shot examples for possible inputs and responses
• Tool use to obtain information, perform actions, often via MCP
• RAG for internal knowledge

The combination of several prompt techniques significantly increases accuracy and thus performance. However, there is of course no miracle cure; the various prompt techniques must be tried out and tested in terms of their performance in specific cases.