Prompt Engineering

Crafting instructions that elicit a desired model outcome without changing weights — the first adaptation lever before heavier techniques like finetuning.

Prompt engineering crafts an instruction that gets a model to generate a desired outcome without updating weights. It is usually the easiest adaptation technique and should be exhausted with experimental rigor before moving to finetuning.

A prompt generally has a prompt anatomy: task description (including role and output format), optional few-shot examples from in-context learning, and the concrete task. Prompting only works if the model can follow instructions; weaker models need more fiddling because they are less robust to small perturbations ("5" vs "five", newlines, capitalization).

Treat prompt changes like ML experiments: version them, track evaluations, and judge each change in the context of the whole system — a subtask win can be a system loss. Separate prompts from application code and prefer a prompt catalog when many apps share prompts. Frame the work as the inbound half of the LLM application loop: stay on the Little Red Riding Hood principle, run a feedforward pass, and gate changes with offline prompt evaluation. Rank harness ambition with prompt engineering sophistication levels before inventing agent loops for problems a thin wrapper or context injection would solve.

---

Sources

• AI Engineering (Chip Huyen), ch. 5
• Prompt Engineering for LLMs (Berryman), ch. 4