Don't Prompt AI. Delegate to It

Back to Devoured - October 23, 2026

DEVOURED

Don't Prompt AI. Delegate to It

The most effective way to use LLMs is to delegate task creation to the AI, moving away from manual prompting toward human-led instruction.

What: The author suggests humans should define 'what good looks like' rather than writing specific prompts, citing research (the 2022 APE paper) where LLM-generated instructions outperformed human ones on 19 of 24 tasks.

Why it matters: This shift treats the human as the product owner or architect rather than a prompt engineer, effectively abstracting away the brittle syntax of LLM interaction.

Takeaway: Instead of writing complex prompts, provide your LLM with a target output example and ask it to write the system instructions necessary to produce that result.

Deep dive

The 'APE' (Automatic Prompt Engineer) paper shows AI-written instructions often outperform human-written prompts.
The process involves multi-round rewrites to enforce constraints like 'do not invent facts'.
Humans should focus on defining evaluative criteria rather than specific syntax.
This approach is specifically effective for automating data-heavy tasks like ticket summarization.

Decoder

APE (Automatic Prompt Engineer): A framework that treats the prompt as an optimization problem, allowing an LLM to generate and iterate its own instructions to maximize success on a given task.

Original article

Prompting is fading as a personal skill, and a weekend-built Jira and Confluence briefing agent shows the alternative: letting AI write the prompt while a human defines what good looks like. Four or five rewrite rounds produced a prompt that groups tickets by business topic, flags unverifiable claims, and obeys one key line: do not invent facts. Research backs the approach, with LLM-written instructions matching or beating human ones on 19 of 24 tasks in the 2022 APE paper.

Explore all 55 articles in today's digest