Big Brain Agent Before and After Context Compaction
Description
A two-panel meme built around an AI coding assistant UI notification. The top panel shows the 'Big Brain Wojak' drawing - a glasses-wearing figure smoking a pipe, seated inside his own enormous wrinkled brain like an armchair, radiating intelligence. The middle is a dark UI banner with a compaction icon reading 'Context automatically compacted', styled like the notice shown by agentic coding tools (e.g., Claude Code) when the conversation history is summarized to fit the context window. The bottom panel is the crude 'Brainlet' wojak - a dented, smooth-headed face with vacant eyes and blue drool - representing the agent after compaction. The joke: an LLM agent is brilliant while it holds full conversation context, but the moment the context is auto-summarized, it forgets critical details and regresses to drooling incompetence, re-asking questions and breaking things it had already fixed
Comments
7Comment deleted
Context compaction is just garbage collection for memories - except it occasionally frees the pointer to why you started the task
Great news: the session fits in context again; bad news: the task does not.
Actually just apposite Comment deleted
I bet your context was compacted. Comment deleted
Codex? Comment deleted
I mean there is a general LLM problem called context rot, so after automatic context compaction output quality becomes better Comment deleted
i don't know why ppl are doing this... it's so much better to task agent with specific instruction and open new session each time you need yet another change Comment deleted