Stop writing LLM instructions in prose: structure criteria as JSON instead

iamrobotbear · x · 2026-09-17

A technical thread on structuring LLM inputs: in @typesafeai's approach, the instructions, criteria and input document can all be JSON — with the first two being the least obvious candidates.

The author notes most people write instructions as prose like instructions: "classify this as a foo or bar" alongside object-form criteria, and argues structuring instructions and criteria reduces parsing ambiguity.

Original post →

More from coding & agent

coding & agent channel →