Prompt Engineering Jobs are a Mirage

‍

‍

There’s a famous saying (misattributed to many), “It’s hard to make predictions, especially about the future.” Ignoring his sage advice, I’m going to make a bold claim, one with little upside if I’m right, and nothing but downside if I’m wrong (legacy evidence on the internet that I was utterly wrong).

Everyone is talking about “query prompting” being the exciting job future. This is a made-up job, or rather one with made-up demand. Understanding why it’s utter nonsense, yet all the rage, can help us identify other illusionary “opportunities” we may encounter in our careers.

If you’ve followed the news in the past year, you know that AI is the hot topic for everyone in the business world. I’ve written repeatedly about it in this blog and spoken about it at events and on podcasts. Time and again I hear “experts” talk about the demand for prompt engineers. To believe the hype, tens of thousands of people will be employed in the coming years. In reality, these jobs will be as in demand as the search engineers were twenty some years ago.

If you don’t know what a search query engineer is, that’s the point. Search engines in the 1990s allowed incredible productivity gains. For example, as a software engineer when I was stuck on some code and my colleagues couldn’t help, I could search the internet and see if anyone posted about this problem and how to solve it. The catch was that search engines weren’t very good back then. The thinking was that companies would need specialists to help do web searches. It wasn’t crazy at the time; for example, law librarians have assisted judges and attorneys with complex research tasks. This was taking that work and just moving it online to the biggest collection of information history had ever seen. That job, if it existed at all, lasted for all of five minutes. Search engines improved, people were not so incompetent, and we’ve been happily web searching ever since.

Today we hear the same siren’s call for prompt engineers. If you search for prompt engineering you’ll find all sorts of articles and other media touting it as the future (the World Economic Forum wrote about it here). If you search the job boards, you’ll find only a handful of roles (I did this search late Sept 2023). Upwork had three jobs. Indeed, when searched with no location, had eight jobs; when the same search on Indeed was done for the bay area it returned one. LinkedIn had thirteen across the US. (Note: I’m only including jobs with prompt engineer as the title. There are plenty of software engineering jobs that include those words as part of a larger description; those are really software jobs not prompt engineering jobs, just as listing “document code” as part of a longer job description doesn’t make the software engineering job a technical writing job.) The website promptjobs.com has a total of eighty-two jobs listed, but most are other types of engineering, and most jobs are from months ago. (Seems like someone saw a quick win opportunity at the start of the summer; we’ll come back to this type of strategy at the end of this article.)

‍

“I just want to say two words to you. Just two words. Are you listening? Prompt engineering.”

‍

Why the disconnect from the hype in the news to the reality of the job posts? Has demand just not yet materialized? In truth, it will never materialize.

In the 1950s you literally needed a PhD to get a computer to do basic math. In the 1970s those with college degrees in software could use languages like COBOL or FORTRAN to do the same, no PhD required. Software and operating systems continued to advance and today a twelve-year-old can get a computer to do a math problem; in fact, today we have twelve-year-olds and younger programming computers to do much more.

The world evolves faster today than it did at the dawn of computing. The clunky search engines of the late 1990s were replaced by better offerings in a matter of years. While demand for AI (and by AI we mean large language models specifically, although AI encompasses many other areas not gaining as much attention in the news) will continue to grow, the interface and usability of the tools is currently growing even faster than the core capabilities of the large language models that sit behind them. While there is some trial and error needed today, and there will always be some, the level of sophistication needed just won’t be high enough to justify a full-time role. There may be some prompt engineers inside AI companies who work as a form of QA (and looking at who is hiring those jobs I saw listed that seems to be many of them), but that’s about it. Prompt engineering jobs are a mirage.

Maybe there will be some companies in the near term that don’t know any better and will hire a handful of these roles. If you get one, and don’t mind working from a company that misguided in their understanding of AI, you’ve got about twelve to eighteen months to jump into a better role before they wake up and realize they don’t need prompt engineers.

This begs the question: if there’s no demand, why is everyone so excited about the role? Understanding the answer can help us avoid similar snake oil roles in the future.

Prompt engineering represents the American dream: a short cut to riches. AI is hot. VCs are throwing billions of dollars at AI companies while management consultants predict hundreds of billions in market opportunity. In turn these companies, tech and otherwise, are throwing big bucks at anyone who can help them win the AI race.

Like computers decades ago, creating these tools requires advanced knowledge. Many of the most in demand people have PhDs or graduate degrees in AI and related fields. There’s demand for software engineers too, but often those with AI/ML (artificial intelligence / machine learning experience). Getting a PhD requires, ugh, hard work. Even if you don’t need a graduate degree, you need extensive computer programming ability, not just a quick boot camp worth of coding that just lets you regurgitate what they taught you in a narrow problem space. But a degree, or enough experience that you really understand advanced programming and basic AI/ML, ugh, that’s also work.

But wait!  Hold the overpriced iPhone you camped out overnight at the Apple store for. There’s a hot new job called “prompt engineering.” It has something to do with AI. Everyone’s talking about it because, unlike search engineering back in the 1990s, today we have tens of thousands of media outlets (websites, blogs, podcasts, social media) and they each have thousands of talking heads who want to sound like they’re cutting edge (he writes as he carefully avoids looking at the mirror). Now everyone can sound as wise and prophetic as Mr. McGuire, “I just want to say two words to you. Just two words. Are you listening? Prompt engineering.”

‍

There’s no one in the world better at prompt engineering than you. Literally, because no one has done it before.

‍

What makes this really great is that there’s no one in the world better at prompt engineering than you. Literally, because no one has done it before. And because there’s no roadmap the bar is low. Unlike the inner workings of an LLM (where even the engineers who made it can’t really explain cause and effect), this is understandable. You just need to repeatedly hit a black box and record the results through trial and error to become an expert. Suddenly anyone who wants to get into AI but doesn’t have time for that, ugh, burdensome training and experience, can jump into the fast-paced, high-paying world of AI. In a year's time you’ll be taking a Blue Origin rocket to party on Necker Island.

It’s like becoming a life coach. <Goes onto LinkedIn and adds “life coach” to title.> Poof, I’m a life coach. Who needs fancy degrees and training! What’s that you said? It’s no longer just a made-up title but one that does have training? You’re right, of course. To be a real life coach I’d have to spend upwards of tens of hours and maybe a few hundred dollars to get my certificate. Why look, I can just sign up for this online course through a local community college! If that sounds like too much, ugh, work, Gallop has a course only four-and-a-half days long (to be fair there are six required practice sessions after, so your training might extend into next week, ugh).

While there are undoubtedly life coaches that do well, most do not. Zip Recruiter reports that the average hourly pay for a life coach in (expensive) New York City is $19.20 an hour and goes on to note that, “regardless of location, there are not many opportunities for increased pay or advancement, even with several years of experience.” A low barrier to entry creates a glut of supply. Even if I’m wrong, and there do turn out to be lots of prompt engineering jobs (I still don’t believe there will be), the pay will be terrible, at best. In other words, don’t become a professional cello player and expect to do as well as Yo-Yo Ma.

We love short cuts to riches and given the riches of AI, shortcuts in this field look like winning lottery tickets. Anytime there is a new field with high demand, there will be a bunch of faux jobs touted by people trying to sound smart, who don’t understand the fundamentals of the industry. History can help us see ahead. Low hanging fruit, particularly in technology, has a shelf life about as long as actual fruit. Consequently, the value of said metaphorical fruit is, again, not unlike that of actual fruit, far lower in value compared to other options.

History also teaches us that if you want to make money during a gold rush, sell shovels. The best way to make money through prompt engineering is to create and sell a certified prompt engineering online course. Caveat emptor!