When building AI systems, make error messages detailed and explanatory, but keep success messages short. AI agents need rich feedback from failures to learn why something went wrong and try again effectively. Verbose successes just waste context.
Be verbose with your errors when working with AI. When you are dealing with AI and agentic processes, what I have found is it tries and then it often fails. And it will really have its success or failure in terms of the whole job to be done based on how it does the retries. And the retries will be best when it has some idea of like why it did something that was wrong. I think often when we make error messages, like with people who are making code, the error our failures are relatively terse and sometimes our successes can be relatively verbose. "Hooray, you're successful!" versus "Boo, that didn't work, but you know, just go ahead and do it again." What we really want to do is the reverse. Your successes should be terse so as not to be consuming all the context and your errors should be verbose. And when you do that, you will find that the AI will try, retry, and succeed.
