Prompt engineer is not a job anymore: which AI skills the market actually buys

In spring 2023 the prompt engineer was the most talked-about job in AI. People asked me where to learn it, and companies wrote "prompt engineer wanted" next to salaries the skill did not justify. Three years later the title has nearly vanished, while the skill stayed and got cheaper. Here is what happened, in numbers, and what the market buys instead.

The rise and fall of the title

The peak is easy to date. According to Indeed, search interest for "prompt engineer" reached 144 per million in April 2023. Today it sits between 20 and 30 per million, several times lower.

The second signal comes from Microsoft. In the Work Trend Index 2024, prompt engineer was among the ten roles leaders planned to hire. In the 2025 edition it is gone from the list.

The third signal is the most interesting. PE Collective, a community of practitioners, reports that the number of roles requiring prompt engineering as a skill tripled between 2024 and 2026, while standalone "Prompt Engineer" titles fell by about 30%. The data is self-reported, but the direction matches every other source.

The skill dissolved into other professions. Much like nobody hired an "Excel specialist" in the 2010s even though every analyst was expected to know Excel.

What it pays

Glassdoor puts the US median for a prompt engineer at 132,949 USD as of May 2026. That sounds fine until you put the neighbours next to it. The AI Engineer median, per LinkedIn data, is 176,000. An agentic engineer, according to the KORE1 survey, earns a base of 155,000 to 210,000 at mid level, 210,000 to 290,000 at senior and 290,000 to 360,000 and above at staff.

KORE1 says it plainly: "prompt engineering alone no longer commands premium pay". The premium comes from RAG, fine-tuning, evaluation and inference skills, which add 15 to 40% on top.

So the market does not pay for writing a prompt. It pays for making the prompt work inside a system that can be measured and maintained.

Russia: the exception that proves the rule

In Russia the title is still growing. According to hh.ru data reported by CNews, prompt engineer vacancies went from 129 in 2024 to 277 in 2025. "Neuro-creator" postings went from 205 to 310.

But they pay under 100,000 RUB a month, while the ML engineer median on Habr Career is around 229,000. The Russian market is repeating the American one with a lag of roughly two years: a spike in the title, then low pay, then dissolution into other roles. I covered the Russian and global numbers in more detail in my review of the AI jobs market in 2026.

What the market buys instead

Look at the 903 US AI Engineer postings Glassdoor analysed in April 2026. The skills they ask for:

• Python: 71%
• PyTorch: 37.7%
• AWS: 32.9%, Azure: 26%
• Fine-tuning: 14.8%
• RAG: 13.6%
• Agents: 10.6%
• Prompt engineering: 8.9%
• Vector databases: 4.5%

Prompt engineering is mentioned less often than RAG, less often than fine-tuning, and almost four times less often than AWS.

In its survey of employers hiring for agentic systems, KORE1 names the number one critical skill: building evaluation harnesses, meaning automated checks that the model answers correctly. After that come tool calling with guardrails and frameworks such as LangGraph, LlamaIndex, AutoGen and CrewAI.

The logic is simple. A prompt can be written in an hour. Proving that it gives the right answer 97% of the time rather than 80%, and that this will not break after the next model update, is engineering work, and that is what gets paid.

What to add if you are good at prompts

If your CV says "prompt engineer", the skill has not gone anywhere, but you need to sell it differently. Four things I would add first.

1. 01Evaluation. Learn to build a test set, run the model through it and compute metrics. This is the KORE1 number one skill, and almost nobody can demonstrate it in an interview.
2. 02Retrieval. RAG appears in 13.6% of AI engineer postings. Understanding how to split documents, how to search them and why the model answers from the wrong chunks is half of real projects.
3. 03Integration. A prompt lives inside a product: APIs, queues, error handling, logs. Python plus one cloud is enough, and AWS is in a third of postings.
4. 04One business domain. Logistics, healthcare, legal, finance. A prompt specialist who understands how a distributor's purchasing department works is worth more than a generalist.

I would change the CV title to AI Engineer or LLM Engineer and leave prompts as one line under skills.

What a business should hire for instead

To owners and hiring managers I say the same thing: do not open a "prompt engineer" vacancy. You will get someone who writes good prompts in a chat window and cannot get a project to production.

Hire for the task. Need AI inside an existing product: look for an AI Engineer with RAG and evaluation experience. Need a process automated with an agent: look for someone who has done tool calling with guardrails and can explain how they check results. Need to figure out where AI will pay off at all: that is a few weeks of a consultant's time, not a permanent position.

A good interview test: ask the candidate to show how they measured quality on their last project. If the answer is "I looked at the outputs", that is a 2023 prompt engineer.

My view: prompts are 10% of the project

In my experience with clients, prompts take up about a tenth of an AI project. The rest is process, data, evaluation and handover to the people who will use it.

An example without names. A distributor wanted an agent that answers stock availability questions. The prompt was ready on day one. Most of the time after that went into getting warehouse data into decent shape, agreeing what the agent does with an ambiguous request, building a test set from real questions and teaching the sales team how to correct mistakes. That does not mean the prompt is unimportant. It means the prompt is not where the project is won or lost.

This is how my services are built: an AI agent for a business process, or AI inside your existing product, with a first working version in two weeks, where the prompt is one part among many. Details are on the services page.

FAQ about the prompt engineer profession

Does prompt engineer still exist as a job?

As a standalone title, barely. It dropped out of the Microsoft Work Trend Index 2025 hiring plans, Indeed search interest fell from 144 per million to between 20 and 30, and standalone postings fell by about 30% according to PE Collective. As a skill it is required three times more often than in 2024.

How much does a prompt engineer earn?

In the US the Glassdoor median is 132,949 USD as of May 2026. An AI Engineer earns a median of 176,000, an agentic engineer 155,000 to 360,000 base. In Russia prompt engineers are paid under 100,000 RUB a month.

What skills should I add to prompting to stay in demand?

Evaluation, retrieval and RAG, integration through Python and one cloud, and one business domain. According to KORE1, RAG, fine-tuning, evaluation and inference skills add 15 to 40% to pay.

Who should a business hire instead of a prompt engineer?

Hire for the task: an AI Engineer to put AI into a product, an agentic systems engineer to automate a process, a consultant to find where AI pays off. In interviews, ask candidates to show how they measured quality.

Why are prompt engineer vacancies growing in Russia?

The Russian market is repeating the American one with a lag. Vacancies went from 129 to 277 in 2025, but pay is under 100,000 RUB, which shows the title sits at the bottom of the market rather than in a growing niche.

If you are planning an AI project and are not sure who to look for, send me a short brief: the task, the data, who will use it. I will tell you whether you need an employee, a contractor or an audit first.