I think that's very informative. Agree with all that. Any tools that you recommend on the context/prompt engineering side with the goals of optimizing the interactions with agents, overall LLM cost, quality and relevance of agent's responses and work output?