Why is teaching employees prompt engineering the wrong approach?

If you are rolling out Copilot or ChatGPT to your company and are trying to teach everyone “prompt engineering” then you are heading for a mess.

Why?

Prompts are critical for accuracy of outcomes. A small, human, variation in prompts will produce huge variations in quality and outcome.

This is just a non-starter for structured workflows.

Effective prompts and instructions sets are non trivial; some of our scenarios run into 100s of lines.

How do you audit all the prompt/outcome pairs for data lineage?

How do you share great prompts across the company so everyone can use them?

How do you link prompts to consistent and structured workflow outputs?

How do you remove prompt engineering in favour of users just typing their ideas in their own natural language?

(And have you looked at how you curate information being used by the models?)

So many problems.

It’s a bit like being at the dawn of the Internet and expecting everyone to learn HTML.

It’s not the thing you need to learn for most people.

If you want to make it easy for your teams then you need to think about the problem differently.

Let’s talk.