Prompt engineering was a transitional skill. It was the assembly language of the LLM era — necessary but temporary. The industry is now moving toward program synthesis: writing structured programs that compile into optimal LLM interactions.
The DSPy Revolution
Stanford's DSPy proved that you can treat LLM calls as differentiable programs, optimizing prompts automatically through compilation. This was the first crack in the prompt engineering paradigm.
Where KARN Fits
KARN takes this further. Instead of optimizing prompts, KARN gives agents a language they can both write and execute efficiently. The agent doesn't need to prompt another agent — it generates KARN code that compiles directly to executable programs.