[dsr_]

Programming is the reification of making decisions -- plus a lot of shiny fashionable chrome. That is, most of it is about thoroughly understanding the meaning of a process, and getting it right.

LLMs don't understand meanings at all. They are anti-compressors. If the information content of a message is how unpredictable it is (and therefore how uncompressable), LLMs generate the most predictable, least information-dense output from your prompt.

So: focus on meaning and decision making. Set LLMs to do fluff and shiny bits.