Prompt Engineering Was Never a Job
We called a temporary keyboard trick a career. The models already took it back.
Two years ago LinkedIn decided prompt engineering was the future. Six-figure salaries. Bootcamps. People who had never shipped a line of code selling $499 courses on "how to talk to AI." I watched grown adults quit real jobs to chase it. It was never a job. It was a bug in the interface, and we mistook it for a profession.
Early language models were powerful and stupid at the same time. They could write a sonnet and then fumble a plain instruction, so you had to coax them. You learned the magic words. "Think step by step." "You are an expert in." Role-play scaffolding, few-shot examples, the whole liturgy. The coaxing worked, and because it worked, people assumed it was a durable skill with a long runway. Durable skills do not get swallowed by the tool itself in eighteen months.
Every model release quietly deletes a chunk of the prompt-engineering canon. The tricks that felt like forbidden knowledge in 2023 are defaults now, baked into the system. You do not tell a good model to think step by step anymore. It already does. The "expert" preamble does close to nothing. The models got better at reading sloppy human intent, which is the entire point of the technology, and the better they got, the less your incantations mattered.
That is not a tragedy. That is the product working as designed.
The grift was betting the gap would stay open. The bootcamp industry needs the gap. The LinkedIn guru needs the gap. A whole micro-economy sprang up to sell you a ladder against a wall that was actively dissolving. They knew, or they should have. You do not build a career on a workaround. You build it on something the tool cannot absorb.
Now, the obvious objection. "But people get paid to build with AI, and that takes skill." Yes. It does. That skill is software engineering, product sense, knowing what to build and how to wire it together and how to make it not fall over in front of real users. Calling that "prompt engineering" is like calling a surgeon a "scalpel operator." The clever sentence you type into a box is the least valuable part of the stack, and it is getting less valuable every quarter.
Here is the part nobody wants to say at the conference. The label was always a status play. "Prompt engineer" let people feel technical without doing the technical work. It was a costume. And costumes are fine until the weather changes, which it has, which is why the job postings that exploded in 2023 quietly thinned out and got renamed to the things they always actually were. Machine learning engineer. Applied AI. Software developer who happens to use the new tools, the same way every developer eventually used Google.
If you spent the last two years getting genuinely good at decomposing problems, evaluating outputs, building systems around unreliable components, and shipping things people use, you are going to be fine. Better than fine. None of that was prompt engineering. That was engineering. If you spent it memorizing magic words, the magic already wore off.
The lesson is bigger than one fake job title. Every wave of this is going to mint a new priesthood that sells access to a temporary gap, and every wave the gap is going to close faster than the last one. The move is not to learn the incantation. The move is to learn the thing underneath that the incantation was a crutch for.
Stop optimizing your spells. The wizard is automating itself.