The Prompt Engineer Who Disappeared, Then Came Back Different

In March 2023, Anthropic posted a job listing that broke the internet a little. The title was Prompt Engineer and Librarian. The salary range was 175,000 to 335,000 dollars a year, more than most senior software roles at the time. Nobody had held this job title two years earlier.

That listing was the opening scene of one of the strangest career stories in modern tech, a role that rocketed up, got flooded by amateurs, nearly vanished, and is now being rebuilt on different foundations entirely.

The Gold Rush

When ChatGPT launched in November 2022, the world discovered that talking to a machine in plain English could replace hours of manual work, provided you knew how to ask. Job postings mentioning GPT rose 51 percent between 2021 and 2022, according to Time magazine. By 2023, LinkedIn saw a 36 fold increase in postings mentioning generative AI compared with the year before.

Forbes ran headline after headline about six figure prompt jobs. Anna Bernstein, a prompt engineer at the New York firm Copy.ai, became a poster child for the role, profiled for writing the instructions that told AI tools how to draft a sales email or a blog post. For a brief window, knowing how to phrase a sentence to an AI model was treated like knowing how to code.

Everyone Became One Overnight

Then the crowd showed up. Bootcamps promised six figure salaries in six weeks. Twitter threads listed 50 prompts that would supposedly change your career. Anyone who had spent an evening with ChatGPT started calling themselves a prompt engineer on LinkedIn.

By mid 2023, Indeed data already showed standalone Prompt Engineer listings starting to plateau. The title was becoming so common and so loosely defined that it stopped meaning much. If a person could get a decent result from ChatGPT after thirty minutes of tinkering, it was hard to argue this was a specialised discipline at all.

Then the Tokens Ran Out

The real reckoning came from research, not opinion. Rick Battle and Teja Gollapudi, researchers at VMware, tested how much popular prompting tricks like chain of thought actually helped language models solve math and logic problems. Their findings were inconsistent. Sometimes a clever prompt helped. Sometimes it made things worse. Much of what passed for prompt engineering was closer to guesswork.

At Sequoia Capital's AI Ascent event in 2024, Andrew Ng, the Stanford professor and cofounder of Coursera, showed the industry something that reframed the conversation. A single prompt to GPT-3.5 solved about 48 percent of a standard coding benchmark, and the newer GPT-4 solved about 67 percent on its own. But when Ng's team wrapped the older, weaker GPT-3.5 in a loop, letting it draft code, test itself, and revise repeatedly, its accuracy jumped to roughly 95 percent, beating GPT-4's single attempt outright.

The lesson landed hard. The model mattered less than the system around it. IEEE Spectrum ran a piece titled AI Prompt Engineering Is Dead. Developer Simon Willison and Shopify's chief executive Tobi Lütke began pushing a new term, context engineering, arguing the real skill was feeding a model the right information, not typing the cleverest sentence. Standalone prompt engineer postings fell sharply through 2024 and 2025 as companies folded the work into broader AI roles.

The Rehiring, on Different Terms

Here is where the story gets interesting again. Companies did not stop needing this skill. They stopped needing amateurs at it. Sloppy, one shot prompting burns compute and tokens at scale, and once businesses were running AI across thousands of daily tasks, that waste showed up directly on the bill.

By 2026, Fortune 500 companies were training existing staff on structured prompting rather than hiring externally, while LinkedIn recorded a 17 percent quarterly rise in postings for agent builder roles. Recruiting data from PE Collective showed prompt related job postings up threefold in 2026, even as the job title itself kept shifting.

In June 2026, Boris Cherny, who leads Claude Code at Anthropic, said publicly that he no longer writes prompts by hand. Instead, he designs loops, repeating cycles where an AI agent acts, checks its own work, and tries again until a goal is met. Peter Steinberger, creator of the agent tool OpenClaw, made the same point around the same time. Andrew Ng picked up the term days later, calling it loop engineering and outlining three nested loops that guide how modern AI products get built.

The pattern is now familiar. Every time AI models get more capable, the skill shifts one level up, from writing a clever sentence, to designing the right context, to architecting the system that lets an AI iterate on its own. What has not changed is that companies keep needing people who understand all three levels.

What This Means for Students

This is exactly the kind of career whiplash that rewards people who understand fundamentals rather than the buzzword of the month. A student who genuinely understands how language models reason, where they fail, and how to build reliable systems around them will always be more employable than someone who memorised fifty prompt templates in 2023 and stopped there.

How Futurowise Can Help

Our Prompt Engineering program is built around this shift. Students learn the craft properly, from writing precise instructions to understanding context windows, evaluation, and the agentic loops now reshaping how companies deploy AI. Instead of chasing a job title that keeps changing, they learn the underlying skill that survives every rename.

Explore the programme: https://www.futurowise.com/programs/prompt-engineering