We were quite certain that prompt engineering was going to be the next high-paying skill because of the emergence of AI and its potential to disrupt many industries.
Your employer was almost certainly going to replace you with someone who could better communicate with AI.
It’s been quite an interesting period and debate because being able to use natural language to get the best possible output from a machine was looking like a skill that could print millionaires.
But is that reality looking less likely given recent advancements in model capabilities?
There’s a distinction between being someone who is great at prompting AI for better outputs and simply being a skilled professional who is tasked with overseeing work done by or with AI.
On one hand, you need to be able to communicate tasks within your skill segment using natural language. On the other hand, you’re mostly confirming the quality of an output, identifying what is wrong, and correcting it.
Understanding this distinction is crucial because people may confuse these two different realities.
Simply because each case has to do with working with AI doesn’t mean they require the same skill.
The first is closer to getting better at communicating with the model itself. The second is more about having enough knowledge of the underlying work to know whether what the model produced actually makes sense.
And that difference becomes increasingly important as models become better at understanding what people mean, even when the instructions aren't perfectly structured.
The reason prompt engineering was even a topic to begin with was because most people aren’t great at communicating.
Quite simply.
Due to this, most inputs are going to be found to be ambiguous, often leading to AI systems taking wild guesses about context and meaning.
A poorly written prompt could produce a poor output because the model had to make too many assumptions about what the user actually wanted.
This created a gap that prompt engineering appeared to fill. If you knew how to structure instructions, provide context, define constraints, and communicate what you wanted more precisely, you could consistently get better results.
But recent models are increasingly capable of handling imperfect instructions, understanding context, and inferring what the user is trying to achieve.
That doesn’t necessarily make clear communication irrelevant. It changes where the value may sit.
Instead of every professional needing to become an expert prompt engineer, the more important skill may increasingly be knowing how to work with AI within your existing area of expertise.
