59 minutes ago · 7 min read1430 words · Tech · hide

I have been using an agent every day for months. Mine is Muse, but almost none of this depends on the brand, it applies to practically any other agent. People ask me how I get it to do useful work instead of producing confident nonsense, and they expect a prompt trick. Photo by Vanishing Point Images on Unsplash. 1. Use specific conversations for specific topics. I keep one conversation for this blog, one for work, one for immigration paperwork, one for each codebase. This looks like tidiness from the outside. The real reason is context. A conversation is the unit of thinking the agent has, and a conversation that mixes your taxes with your screenplay gives you an agent that is slightly wrong about both. Separate the topics and each thread grows its own history: what was already decided, which words mean what, who is who. Most of what people call prompt engineering is really this, deciding what the agent is allowed to be thinking about. 2. Set schedules to compact your conversations.…

Commenting on Bubbles has been removed. Why? If you want to talk about this post, write to the author, use the comments on their blog if it has them, or reply with a "RE:" post on your own blog. Suggestions? feedback@bubbles.town