Originally titled "Where are all the prompt engineers?" until I finished writing and realized I was connecting a lot more dots.

Via Fast Company:

Part of the prompt engineer’s appeal was its low barrier to entry. The role required little technical expertise, making it an accessible path for those eager to join a booming market. But because the position was so generalized, it was also easily replaced.

Those easy-access AI jobs may no longer exist. Machine learning engineering roles demand deep technical expertise—skills that take years to develop, unlike the relatively shallow learning curve for prompt engineering. Even basic coding skills are no longer sufficient. Indeed’s Shrivastava notes that while demand for developers is declining, engineering roles more broadly are on the rise.

I remember the blogs, Substack (blech) posts, and LinkedIn musings on prompt engineering. People were so excited, and to be honest, I get it—if I wasn't already in tech, it might seem like a way to make a transition. It was really accessible... until the other shoe dropped.

AI companies will always want so-called AI to be as accessible as possible. Their business models absolutely depend on it. OpenAI can't possibly justify their valuation unless absolutely everyone is using it. Employers would be happy if they no longer needed to hire workers with specialized skills, so the specialization that prompt engineering offered was never long for this world.

It remains unclear whether companies were ever truly hiring for individually titled prompt engineers. They certainly aren’t now, says Allison Shrivastava, an economist with the Indeed Hiring Lab.

I never once saw a role posted for a prompt engineer. In most engineering departments I am familiar with, a call for a prompt engineer would be seen as unserious, not because of the candidates but because even early on it was seen as an easy skill to acquire rather than the basis for a career. If you had two candidates for a prompt engineer role and one knew backend development while the other did not, you'd hire the backend developer every time because that's the much more important skill.

This reminds me of There is no Vibe Engineering (worth a read). Writing code or getting the LLM to give you the result you want is only a small portion of the value you get from an engineer. Knowing what code you actually want to write is the real value prop. A good prompt engineer isn't really a prompt engineer: they're a developer who can also talk to an LLM, and the people who make LLMs want everyone to be able to talk to LLMs.

“I wouldn’t say that [there are] new jobs, necessarily; it’s more so that it’s changing how people work,” [Tim] Tully says. “You’re using AI all the time now, whether you like it or not, and it’s accelerating what you do.” 

In other words, the Jevons Paradox but for labor. Many workplaces are asking their workers to use AI tools to produce more widgets faster (doing more with less), not because they want to give people 4-day workweeks and European vacations, but because they want to increase their profit margins. They don't want to miss this bus (like they didn't want to miss blockchain) and capitalism is a zero-sum game of winners and losers.

Keynes predicted we'd all be working 15 hours a week by now, and we probably could be, except he forgot the human factor. If the bosses can squeeze more juice from the lemon, they will. Even if the lemon is a human being.