It is programming the AI, and most companies should look for it among their own experts. Prompt engineering is often mistaken for typing commands into a chat.

The real work of prompt engineering starts before the prompt is written, when someone decides what the AI should achieve. The work continues after the AI answers, when someone checks whether the AI achieved it. So most companies should look for prompt engineering skills in the people they hire for other jobs, because few need a separate prompt engineer.

Castells saw the split coming

Handing work to machines was foreseen long before generative AI. Sociologist Manuel Castells divided work into:

  • self-programmable labour, which works out on its own how to reach a goal, and
  • generic labour, which carries out set tasks and is eventually replaced by machines.1

The split is now visible in the labour market.

In jobs exposed to AI, young workers are losing ground while experienced workers are gaining2.

The likely reason is what each group knows. Young workers bring mostly what can be learned from books and written procedures, and AI has that knowledge too. Experienced workers bring know-how nobody has written down.”

This also links back to Castells’ ideas. In short: what can be written down goes to machines.

A common skill, a rare job title

For everyone who works with AI, prompting may become as basic as reading and writing.4 Yet the job title stays rare. In January 2026, 7,359 US job postings mentioned prompt engineering, while 5 of the 66,785 CVs in one recruitment database used the title.5

Writing prompts does not make anyone a prompt engineer.

Where a prompt engineer is needed

A prompt engineer is needed where the same prompts run again and again in products and processes, and nobody reads each answer. In such products and processes, one error repeats at scale, so someone has to check the answers before the AI goes to work. These people write the standing instructions the AI follows and test how it behaves.6,7 By contrast, a one-off question in a chat needs no prompt engineer. In ordinary jobs, checking AI’s answers falls to the people who use AI, so AI makes their work even more demanding.8

Plain language sets tasks, not programs

AI now handles plain language so well that prompt engineering may seem unnecessary.

Plain language is enough to set a task. Programming needs a precise specification, and plain language leaves too much room for interpretation.9

Moreover, the instructions and the documents AI works on are written in the same plain language, so AI cannot always tell which words are instructions.

The UK’s cyber security agency warns that attacks hidden in such documents may never be fully stopped.10

So typing commands into a chat is not prompt engineering.

What makes prompting engineering

An engineer designs a solution to defined requirements and evaluates it against them,11 the way a bridge is designed for a given load and calculated to carry it. So prompting becomes engineering only when someone has defined in advance what a correct result is,12 and every change is tested.13

Prompt engineering, also called promptware engineering, is in fact a framework for using prompt architecture efficiently.14

How prompting becomes programming

Researchers already treat some prompts that run inside software as programs.15 Writing such a program comes down to the result it must produce and the tests that check it, more than to the choice of words. Unlike an ordinary program, AI may give different answers to the same prompt,16 so only tests show whether the results are right.

What to look for in a prompt engineer

So a company hiring a prompt engineer should test not how well candidates write prompts but whether they can define a correct result and check it. The candidate also has to know the field the AI works in, because only such a person can tell a good answer from a bad one.

In coding with AI agents, sessions of users rated as experts reach verified results more than twice as often as sessions of beginners.17

People who lean on AI can lose skills they no longer practise, such as detecting problems on their own,18 so the candidate should show they can judge results without AI.

The prompt engineer most companies already employ

For most companies, prompt engineering therefore needs no separate hire.

It is a skill to develop in the experts who set up AI to run the company’s processes.

These experts already know what a good result looks like. But to work with AI, they have to become engineers too. So prompt engineering is not a new profession. It is the new way an expert does their job.

Sources:

  1. Edward Elgar via USC Annenberg, Informationalism, Networks, and the Network Society: A Theoretical Blueprint, 2004.
  2. Stanford Digital Economy Lab, No Widespread Displacement, but the AI Employment Gap for Young Workers Has Widened to 19%, 2026.
  3. arXiv (preprint), A theory-based AI automation exposure index: Applying Moravec’s Paradox to the US labor market, 2025.
  4. US News via Vanderbilt University, How to become a prompt engineer, 2026.
  5. RezScore (vendor), Prompt engineering jobs in 2026: skill yes, title no, 2026.
  6. Anthropic job posting via Accel, Prompt Engineer, Claude Code, 2026.
  7. Appian job posting via ResumeGeni, AI Prompt Engineer, 2026.
  8. Simon Willison on a Harvard Business Review study, AI Doesn’t Reduce Work, It Intensifies It, 2026.
  9. arXiv (preprint), Specifications: The Missing Link to Making the Development of LLM Systems an Engineering Discipline, 2024.
  10. NCSC, Mistaking AI vulnerability could lead to large-scale breaches, NCSC warns, 2025.
  11. ABET, Criteria for Accrediting Engineering Programs 2026-2027, 2026.
  12. Anthropic (company documentation), Define success criteria and build evaluations, 2026.
  13. OpenAI (company documentation), Evaluation best practices, 2026.
  14. arXiv (preprint), Promptware Engineering: Software Engineering for LLM Prompt Development, 2025.
  15. arXiv (preprint), Prompts Are Programs Too! Understanding How Developers Build Software Containing Prompts, 2024.
  16. arXiv (preprint), Stochastic CHAOS: Why Deterministic Inference Kills, and Distributional Variability Is the Heartbeat of Artifical Cognition, 2026.
  17. Anthropic (company research), Agentic coding and persistent returns to expertise, 2026.
  18. Scientific American, Is AI ruining our skills? Early results are in, and they’re not good, 2026.