Weird New Job Alert: What Is an AI Prompt Engineer?

Did you see the headline(Opens in a new window) about an AI job that pays $335,000 per year without the need for a computer engineering background? It’s no joke: The role of “AI prompt engineer” is a real profession. But while it would be amazing to make that kind of money in an emerging field with little to no experience, it’s not that simple.

Generative artificial intelligence (AI) tools—including ChatGPT, Google’s Bard, Midjourney, Dall-E, and others—require that users speak their language. Ostensibly, that language is natural-sounding, everyday English (or whatever other human languages your chatbot of choice supports). The text fed into the AI to elicit a response is called a prompt. 

The whole point of a prompt is to take advantage of natural language processing (NLP), which lets you ask an AI a question using normal words and syntax as you would a real person. There’s no programming involved. 

Prompt engineering—some call it “AI whispering”—points to the limitations of NLP. It’s not enough to toss off an idea and see whether the AI can deliver. Knowing how to write a decent query for individual AI tools is a major factor in getting the best results from and better training for the AI. But as experts note below, it may not be necessary to be a “whisperer” forever.

What Is a Prompt? 

(Credit: Tada Images/Shutterstock)

When you want something from a generative AI, it’s not going to intuit an answer. These tools aren’t truly intelligent and definitely aren’t sentient. You have to ask the chatbot for what you want. 

You do that with a prompt: a string of words that spell out what you seek. The broader your request, the broader the AI’s interpretation might be. Asking for “dogs painting” or “tell me a joke” can bring up some weird, possibly useless responses. But requesting “a St. Bernard dog wearing a beret while painting the Eiffel Tower with oil paints at an easel, in the style of Vincent Van Gogh” or “a joke about Vladimir Putin in the style of Rodney Dangerfield” narrows potential output. Specifics matter. 

Creating an AI prompt is more of an art than a science. But it’s an art you need to practice. In an interview with Forbes(Opens in a new window), the CEO of China’s Baidu search engine, which makes the Ernie AI, claims that in a decade, “half of the world’s jobs will be in prompt engineering” and everyone else will be obsolete. 

On the other hand, Sarah Shaiq, former chief product officer at 3DLOOK, says that prompt engineering is having a moment now only because of the limitations of the current GPT architecture. “As the architecture evolves, the need for prompt engineering will get absorbed into the standard educational path on how to use the system,” Shaiq told us.

What Does a Prompt Engineer Do?

AI prompt engineering can mean one of two things: Either you’re writing prompts to get stellar results from an AI (text or art that can be used elsewhere, for example), or you’re writing copy to train and test an AI system. 

Of course, AI can be biased, generate misinformation, and produce hallucinations. That’s not necessarily the AI’s fault—it is, after all, not really thinking. AIs are simply incredible mimics and remixers, taking the billions of parameters they’re trained with and spitting out something that approximates human-generated content. A skillful prompt engineer can mitigate any problems and improve the output. 

Prompts can and should be revamped and rewritten constantly to perfect the outcome. That’s part of what a prompt engineer does: Create a quality prompt that results in high-quality output.

What Skills Does an AI Prompt Engineer Need?

(Credit: salarko/shutterstock)

To succeed in prompt engineering, you must be good with words. You essentially need to be a writer to generate AI writing—or AI anything. The secret is to include the right specific details. Being ambiguous will lead to results you don’t want—although it can also create serendipitous effects you didn’t know you wanted until you saw them. It’s a high-wire act. 

The need for specifics is slightly mitigated by the fact that some AIs are all about having a full conversation, allowing for “chain-of-thought prompting(Opens in a new window)”: The AI shows its reasons for creating certain output, and you refine the prompt as the two of you “chat.” Things will change even more as the systems start to remember your preferences and previous conversations, Shaiq notes. Then, the systems may auto-prompt for you.

A skilled prompt engineer learns the shortcuts necessary to get the most out of an AI. For example, perhaps you want a prompt to be open to interpretation by the AI. You could ask a tool like Midjourney to generate art based on a vague idea, using an indistinct prompt. But you could also refine a prompt(Opens in a new window) in ways that are not natural language. Midjourney supports special codes that can set the aspect ratio for art or that give weight to styles you mention in the prompt. 

The same is true for other generative AIs, and prompt engineers should know about these kinds of extra options. 

Where Can I Learn AI Prompt Engineering?

It’s a new field, but the buzz is tremendous. On YouTube, you can already find thousands of hours of prompt tutorials. Online classes such as the $99 PromptHero Crash Course(Opens in a new window) and the free guides at Learn Prompting(Opens in a new window) are multiplying. Online education sites, including Udemy(Opens in a new window) and Coursera(Opens in a new window), have also jumped on the prompt-engineering tutorial bandwagon.

OpenAI, which created ChatGPT, has launched a free (for now) one-hour course with DeepLearning.AI(Opens in a new window) to get people started. 

If you prefer to learn by example, you can check out repositories of pre-made prompts. The aforementioned PromptHero is one; it displays AI art with the prompt that created it. The site also has ChatGPT prompts. All its prompts are rated by other users, to help you decide which to try. 

Similar lists can be found at FlowGPT(Opens in a new window). A tool such as Pic2Prompt generates a prompt based on an image you provide. You’ll find a never-ending list of prompts for ChatGPT at Prompts.chat(Opens in a new window). And for those willing to pay, marketplaces such as PromptBase(Opens in a new window) and Prompt.AI(Opens in a new window) sell prompts created by someone else. Or you can sell your own. There are and will be many more such services: Everyone wants to ride the AI gravy train. 

One interesting way to learn how to write AI prompts is to use an AI. Many people turn to ChatGPT to ask it to generate a prompt(Opens in a new window) that will work with the art generators. And you can request a fully optimized, rewritten prompt for almost any generative AI using a new browser extension called PromptPerfect(Opens in a new window). It is one of the first wave of plug-ins using ChatGPT+.

As Shaiq notes: “The AI ecosystem is already working on automating prompt engineering.” Knowing all the ins-and-outs may not be as necessary in the future.

How Many Prompt Engineer Jobs Are Available Now?

As of this writing, a search on Indeed.com for “prompt engineer” brings up 956 jobs across the US—and not all of those are truly related to generative AI. Only 164 pay over $125,000 a year; few pay more than $200,000, and I found only one that topped $300,000. 

Still, the current market for prompt engineers could be a bubble(Opens in a new window) waiting to pop. But AI is not going away. It has supplanted the overblown metaverse and VR as the next big thing. The possibilities of getting in early by mastering the ways of the prompt could be endless if the evolution of (and the hype for) AI persists. Give yourself a few hours to try creating some stellar output with your own prompts, and who knows where you might land. 

Just keep in mind, as Shaiq says, that all this learning may soon not be necessary: “I don’t believe that this is going to be a specialized, independent skill set in the future.”