I’m already on it lol

@Jon Fritsch 5 minutes of Fable and 5 hours of wait. Isn't that wonderful?

That’s most likely due to context bloat. When you first started, your agent probably had very little context, so it wasn’t filling up the context window very quickly. Over time, however, things accumulate. Your episodic, semantic, and procedural memories grow, and much of that gets injected into the system prompt or retrieved as additional context every time you interact with the agent. As the context window becomes increasingly saturated, agents tend to produce lower-quality outputs (“slop”) and take longer to process requests because they have significantly more information to reason over. This is why context engineering is so important. The goal is to provide the right context, at the right time, and in the right amount—not simply more context. I covered several context engineering techniques, including using folder structures to organize and retrieve context efficiently, in one of my recorded videos if you’d like to check it out. I’d also recommend building a memory wiki in Obsidian. It allows you to organize and index your knowledge base so agents can retrieve only the relevant information instead of ingesting large amounts of unnecessary context. That keeps prompts leaner, improves response quality, and helps make much better use of your available context window.

@Jonathan Proulx Navigate to the Classroom and checkout the local LLMs segment under the courses.

Sign me up, buddy!

This didn't age well, did it? haha