What new AI jobs are appearing?

Every big technology shift creates jobs that did not have names before. Large language models, the systems behind ChatGPT and Claude, are doing the same, and some of the new titles already appear in real postings.

The roles below are worth understanding even if you never apply for one, because pieces of them are spreading into ordinary jobs: writing good instructions for an AI, judging its answers, and thinking about who could be harmed by how it is used.

Prompt engineer

The most talked-about listing this year came from Anthropic, the San Francisco company that makes the Claude assistant. In March, Fortune reported its posting for a "prompt engineer and librarian," with a listed salary range of $175,000 to $335,000. That figure is from one well-funded startup and is not typical of the field.

The job, as the listing described it, is to work out the best ways of prompting the company's AI for many kinds of tasks, then document them as a library of prompts and tutorials so others can learn. Anthropic acknowledged the role was hard to fill, since prompt engineering as a field was under two years old. Instead of asking for a particular degree, it asked candidates to show that they could get complex behavior out of Claude through a well-designed series of prompts.

AI trainer

Chat assistants are not only trained on internet text. To make them follow instructions, companies hire people to show them what good answers look like. OpenAI described the method in early 2022 when it introduced its InstructGPT models. Human labelers wrote example answers to prompts, then ranked several answers from the model from best to worst. Those rankings trained a separate model to predict what people prefer, and the language model was then tuned to score well on it. The technique is called reinforcement learning from human feedback, or RLHF.

The people doing this work go by titles such as AI trainer or labeler. Their judgments matter: OpenAI noted that the model ends up reflecting the values of the people who labeled its data.

AI ethicist and responsible-AI specialist

Large companies have built teams to anticipate how AI products might cause harm. A Verge and Platformer report in March described Microsoft's ethics and society team, which at its peak in 2020 had about 30 people, including engineers, designers and philosophers. Its job was to make sure the company's responsible-AI principles actually showed up in product design. It built a role-playing game called Judgment Call to help designers imagine harms, and it wrote a memo on the risks of the Bing Image Creator before that tool launched.

The same report also shows the fragility of these jobs: Microsoft laid off that team in March as part of 10,000 job cuts, while keeping a separate Office of Responsible AI that writes rules and principles. Anyone considering this path should know that the work is valued in principle and sometimes cut under pressure to ship.

A day in the life, side by side

Here is how a typical day might look in each role. The schedules are illustrative, built from the duties described in the sources above, not from a single real person's diary.

Time	Prompt engineer	AI trainer	Responsible-AI specialist
Morning	Tests five ways of asking the model to summarize contracts and scores each against a checklist.	Writes model answers to a batch of user prompts, following a style guide.	Reviews a new feature's design and lists ways it could be misused.
Midday	Documents the best prompt in the shared library with examples and known failures.	Ranks four model answers per prompt from best to worst and notes why.	Runs a workshop where designers play through harm scenarios.
Afternoon	Meets a customer team to adapt prompts to their documents.	Flags prompts where the guidelines are unclear for the team lead.	Writes a short risk memo recommending changes before launch.

Try it: Pick one of the three columns and do a 30-minute version yourself. For example, ask a chatbot the same question four different ways and rank the answers from best to worst, writing one sentence on why. That is a small taste of both prompt engineering and AI training.

What do AI trainers, ethicists and prompt engineers actually do?

In short: prompt engineers find out how to ask, AI trainers teach models what a good answer looks like, and ethicists ask who might be hurt and how to prevent it.

Prompt engineers spend their time experimenting with wording, examples and structure, measuring which version gives reliable results, and writing it down so others can reuse it. The skills are clear writing, patience with testing, and enough knowledge of a domain to judge results.

AI trainers produce the human judgments models learn from. They write demonstration answers, compare model outputs and follow detailed guidelines, often at volume. The skills are careful reading, consistency and, for specialist work, subject expertise.

Ethics and responsible-AI specialists sit between product teams and the people affected by products. They review designs, run exercises to imagine misuse, write policies and recommend changes before launch. The skills are a mix of technical understanding, social science or philosophy, and the ability to persuade teams under deadline pressure.