Back in January, Andrej Karpathy, who now works for OpenAI LP and used to be the director of artificial intelligence for Tesla Inc., tweeted: “The hottest new programming language is English.”

Karpathy was only semiserious, yet he has identified a new career path: AI chatbot prompt engineer. It could catch on.

The term describes the people who create and refine the text prompts that users type into the chatbot query windows — hence the use of English, or any other standard human language. These types of engineers don’t need to learn any code, but they do need to learn how the AI chatbots work, what they’re good at doing and what they’re not good at doing.

Think of it as Ann Landers giving advice but for lonely computers instead of humans. Karpathy has called them “AI psychologists.” Alex Furmansky has already created one called Esther Perel, the therapy bot. (You text “her” your questions.)

I asked a therapist friend what she thought of the idea, and she liked the concept. “I can imagine those who need to speak to someone urgently, and Esther can be there for them, always with a positive or neutral response, especially questions, which is what we do in order to have people find their own answers.” Some of us are old enough to recall the first experiment in this genre, Eliza, which is still running.

Consultant Shelly Palmer calls this skill prompt crafting and claims that “it will be as important from a productivity standpoint as learning to write prose, doing arithmetic, or sending an email.” Moreover, the skill may prove critical to companies looking to leverage generative AI. Tim Barkow, a product designer at numerous tech companies, including IBM Corp., recently wrote on LinkedIn: “Ensuring that your AI models are contributing positively to your business is going to take some effort to get right.”

Professor Ethan Mollick at the University of Pennsylvania’s Wharton School told Time magazine he agreed with this cautionary perspective. “We don’t know if there’s a special skill involved for prompt engineering or if it just requires a lot of time spent with chatbots,” he said.

To that last point, consider Jason Allen, the artist who won a Colorado State Fair competition last year with artwork created by AI-based Midjourney Inc. He said he took more than 80 hours and 900 prompts to get the winning picture created. He has refused to reveal the final prompt sequence, which illustrates something. And there is a startup called PromptBase.com where people can buy and sell their own prompts to generate someone else’s art.

Another data point: The popular website builder Wix.com Inc. now can create an entire site from a series of prompted responses with its AI Site Generator tool, launched this week.

But let’s look at a company that has gone all-in on AI, the global retailer Newegg Inc. It claims to be the first to offer a ChatGPT browser plug-in to assist shoppers in building PCs from a collection of components.

Lucy Huo, Newegg’s vice president of application development, who has been working with AI tools for five months, told SiliconANGLE in an email that the company doesn’t use any specific prompt engineering tools, other than continuous testing and trial and error refinement, along with plenty of customer feedback. Newegg doesn’t have any prompt engineers but relies on their regular software engineers to develop the needed expertise.

Riley Goodside was one of the first, albeit self-proclaimed, prompt engineers and started explaining how he interacts with the chatbots in this Twitter thread back last October. One of his maxims was to figure out when and why the AI gets things wrong or produces internet citations that are completely bogus.

That became an issue when two New York City lawyers were fined for producing a legal brief with erroneous citations. Perhaps they should have hired their own prompt engineering talent or taken one of Rob Lennon’s online classes on how to do the job. Speaking of online training, Amazon Web Services Inc. has created a series of free online classes, including one that’s an introductory coding course for developers using its Amazon CodeWhisperer.

Goodside was first interviewed in February for the Washington Post, then in more detail for Semafor earlier this month. He now spends his time thinking about “adversarial prompting,” meaning ways he can game the prompts to get the best – and hopefully accurate – results. He’s bullish on prompt engineers who are just “ordinary people” who will be able to get results that would have been difficult to do before AI models became popular.

I checked the job boards for prompt engineering openings and wasn’t impressed. On Indeed.com, there were only 13 open positions. On LinkedIn there were 282 listed, with most from head hunting firm Crossover. If you can believe the posted salaries, most were in the low six figures. At the lower end, freelance site Fiverr lists 367 providers, starting at $5 per job, as seen in this screen capture:

Anthropic, a Google LLC-backed AI startup, has had this job posted for months for a “prompt engineer and librarian” in San Francisco. With a salary range of $250,000 to $335,000, that might be interesting, until the posting asks for five years “relevant experience,” whatever that means.

So it may be true that prompt engineers are still more of a sideline than a career path. It’s hard to tell, because elements of the job are clearly important for the future of our chatbot overlords.

Jim Louderback, who is editor and publisher of the newsletter Inside the Creator Economy, has been happy with the results of using an AI-based headline generator for his newsletter. “People are still learning how to use the tool, I don’t think it is yet a discrete job,” he told SiliconANGLE. “But the average worker needs to be conversant and speaking the prompt lingo.”

Louderback compares this to how we all learned basic grammar and punctuation back in elementary school. He recommends that everyone should uplevel their skills, “because the chatbots still require humans to sit in the middle to determine what output is accurate.” This is the Newegg model.

Even if its popularity increases, the prompt engineer could be subject to the same job insecurity that threatens traditional coding engineers, and the role could be replaced with an AI-based tool. An early version of this can be found with an open-source project by Matt Shumer. There is some delicious irony to that situation.

Hany Farid, a computer science professor at University of California at Berkeley, told SiliconANGLE that “it can be difficult to separate the hype from the substance in this latest AI revolution. I am, however, skeptical that prompt engineer will remain a viable career path. First, we shouldn’t call it engineering. What is being described is more akin to ‘prompt tweaking’ to manipulate around the quirks of LLMs and generative-AI. Second, as these systems are perfected, many of these quirks will vanish and the systems will operate as we would want them to, simply understanding what it is that we are asking of them.” Both are good points.

“The complexity of the things that we demand of our language models will grow,” Goodside told Semafor. “Prompt engineering will shift to a new frontier. You’ll still have prompting in some sense, but you’ll be prompting at a higher level and asking it to do more impressive things.”

Another optimist is Latent.Space’s Shawn Wang, who is running an AI engineering conference in San Francisco in the fall. “You do not start driving by reading the schematics for the Ford Model T,” he writes. “As human engineers learn to harness AI, they will increasingly do engineering as well, until a distant future when we look up one day and can no longer tell the difference.”

Writers still could be the next endangered species. Google has been touting its AI news story tool called Genesis at various newsrooms. In the meantime, I can guarantee that I wrote every word in this article — from scratch, with manual searches.

Image: Fiverr/Pixabay