Prompting is not a career, because it is becoming a general skill like typing or searching: useful to everyone, distinctive to no one. Supervising AI agents that act is different. It carries accountability for decisions with real consequences, it needs domain judgment and it is checked. That is the kind of work that forms a career.
Why is prompting not enough to build a career on?
Three reasons.
Everyone is learning it. Writing clear instructions to an AI model is now taught widely and practised daily in ordinary office jobs. When a skill becomes general, it stops being a reason to hire someone in particular.
The tools are absorbing it. Systems increasingly help users phrase requests, suggest instructions and assemble briefs from plain descriptions. In Pryme, for example, Agent Studio lets a person describe the job in plain language and assembles the agent's brief from it. The careful wording that was once a craft is becoming part of the product.
Prompting does not carry accountability. A good prompt produces a better draft. But in a business that runs agents that act, the question that matters is who decides whether the action goes ahead. That is not a prompting question.
None of this means prompting is useless. It is a good skill to have. It is just not a profession.
Compare it with spreadsheets. Knowing formulas once set people apart; now it is expected, and the careers built on spreadsheets belong to the analysts and accountants who decide what the numbers mean. Prompting is heading the same way.
What does supervising agents involve?
Supervising governed agents is accountable work. An agent handles routine cases within its bounds and holds the rest for a named person. The supervisor decides those held items: approves, returns with a reason or escalates. Each decision is filed in an append-only record with their name.
Beyond the queue, supervision includes reading the workspace to see which items are waiting too long, noticing when an agent's output starts to drift, knowing what happens when an agent's certificate lapses and it drops to read-only, and keeping the agent's sources current when procedures change. It means understanding depth, which sets how far an agent may act on each capability, and verdicts: held, refused and withheld.
That combination of judgment, domain knowledge and accountability is what employers struggle to replace, and it does not get absorbed into the tools. The more agents act, the more it matters.
Why will prompting courses not prepare you?
Because they teach a different task. A prompting course teaches you to get better output from a model. Supervising agents teaches you to decide whether an agent's proposed action should happen, given its bounds, its sources and the business's rules.
The two can look similar from a distance; both involve AI and both involve instructions. Up close they are different jobs. A prompt engineer asks, "How do I get a better answer?" A supervisor asks, "Is this agent allowed to do this, on this evidence, for this customer, right now?" The page on why prompting courses will not prepare your team for agents that act makes the argument from the employer's side.
What skills does supervision need instead?
Mostly skills you may already have from your current work.
- Attention to detail. Spotting the wrong date, the mismatched amount, the instruction hidden in an attachment. The page on what attention to detail means is a useful check on how you describe it.
- Weighing evidence by source. Knowing that an emailed claim is weaker than a record from the system of record.
- Knowing the red lines in your field and never crossing them.
- Writing reasons others can follow.
- Escalating well: neither burying problems nor passing on everything.
None needs programming. All need practice on realistic cases.
What is the trade-off?
Supervision is heavier work than prompting. Your name is on decisions. Mistakes have consequences for customers and the business. Some people prefer the lighter responsibility of producing drafts for others to decide.
Supervision is also less glamorous. It does not appear in headlines about AI. It looks like operations work, because it is. For people who want an AI career that sounds technical, that may feel like a step down. For people who want work that lasts, it is the more solid choice.
And building agents, the step beyond supervision, does involve writing briefs, which is a close relative of prompting. If you enjoy that side, supervision is still the right starting point: builders who have never supervised design agents that look good in testing and frustrate reviewers in production.
Can you train for it while working full-time?
Yes, and many people will have to. Supervision training is measured in hours and weeks, not years, and much of it can be done around a job. The pages on training while working full-time in Denmark and in Switzerland look at two markets. A structured approach is set out in a 90-day AI skills plan for career changers.
What should you do differently?
If you have spent time on prompting courses, keep the skill, but stop there. Put your next effort into learning how governed agents decide, practising review on realistic cases and getting evidence an employer can check, such as a certificate with a practicum. Describe yourself as someone who can supervise agents, not someone who can talk to them. The overview of AI skills for job seekers sets out what employers will ask you to prove.
Questions people ask
Is prompt engineering still a job title?
Some employers still use it, often for roles that involve building and testing agents. Read the duties: if the work includes deciding what agents may do and reviewing their actions, it is closer to supervision than the title suggests.
Do supervisors need to write prompts at all?
Rarely in the sense of crafting instructions. They do write reasons and propose changes to an agent's brief or sources, which needs clear writing.
Will supervising agents also be automated?
Parts of it, such as routing and sorting, may be. The decision on a held item stays with a person by design, because someone must answer for it.