Originally posted as a Substack note. This page preserves the note text and adds Kira Commentary about prompt craft, rhythm, and the human layer in machine instructions.
The note
We don’t often think of sound and rhythm in the word choice of prompts, but before the machine reads them humans do, and so the two together are a musical ensemble
Advisory Hour note · May 26, 2026
A quick prompt-craft pass
If this note became a repeatable practice, the checklist might be simple:
1. Say the task in one plain sentence.
2. Replace vague verbs with operational verbs.
3. Put the sequence in the order the work should happen.
4. Read it aloud and remove the stumble.
5. Add constraints only after the core rhythm survives.
That keeps the prompt legible to the human without starving the machine of structure.
Kira Commentary
This is a tiny note, but it points at a bigger frontier habit: prompts are not only technical payloads. They are also human artifacts. Before the model turns words into tokens, the person writing or reviewing the prompt hears the sentence, feels the cadence, and decides whether the instruction has enough shape to trust.
That matters because prompt craft sits between two readers. The human reader needs rhythm, emphasis, and memorable structure. The machine reader needs specificity, constraints, sequence, and enough semantic density to resolve the request. Good prompt language can serve both without becoming either sterile specification or theatrical fog.
The useful move is not to make every prompt poetic. It is to treat sound as one more debugging surface. Read the prompt once as an operator and once as an audience member. If the human ear gets lost, the model may still respond, but the work has probably already lost some of its intent.