Why Claude's output feels cognitively expensive:

Claude structures sentences backward from human speech patterns. It leads with the conclusion, forcing your brain to buffer context it hasn't received yet.

Example breakdown:
"A missing semicolon caused the crash" - subject is the detail you're still parsing. Your working memory has to hold "missing semicolon" while waiting for "crash" to give it meaning.

Human-optimized version: "The crash came from a missing semicolon" - starts with the known entity (crash), adds the new info (semicolon) at the end. Zero cognitive overhead.

Claude's own explanation: "I write from where I ended up. A person talks from where the listener is standing."

The fix: When using LLM output, rewrite sentences to put familiar concepts first, new information last. Saves brain cycles, reduces that melting feeling.