Every week I see another LinkedIn post from someone calling themselves a "prompt engineer" like it is a career path. It is not. It is a skill. An important one, sure. But building a role around it is like hiring a "Google search specialist" in 2005. The real jobs are elsewhere, and we are wasting time pretending otherwise.
The Hype Cycle Did This
When ChatGPT landed, everyone rushed to figure out how to talk to it properly. Fair enough. The difference between a bad prompt and a good one is enormous. People who understood how to structure inputs, chain reasoning, and extract consistent outputs had a genuine advantage.
Then the job postings started. "Prompt Engineer - $150K USD." Companies were hiring people whose entire role was writing prompts. And look, in early 2024 that might have made sense. The tooling was primitive. The models were fussy. Getting reliable outputs required real craft.
But here is the thing. The models are getting better at understanding bad prompts. Every major model update makes prompt sensitivity less of an issue. The gap between a carefully crafted prompt and a straightforward one is shrinking with every release.
If your entire value proposition is knowing how to talk to a model, you are one model update away from irrelevance.
Mak Khan
Chief AI Officer
What the Real Jobs Are
The skills that actually matter in enterprise AI are not about writing prompts. They are about everything around the prompts.
Architecture. How do you design a system where AI components talk to enterprise systems? How do you handle model routing, fallback logic, context management, and token budgets? How do you build a pipeline that works at scale and does not fall over when the model provider has an outage? That is architecture. That is a career.
Evaluation. How do you know the AI is working? Not "does it feel right" but actually working, measurably, consistently, across edge cases. Building evaluation frameworks, defining metrics that matter, running regression tests against model updates. That is evaluation engineering. That is a career.
Operations. How do you run AI in production? Monitoring drift, managing costs, handling model updates, maintaining guardrails, responding to failures. AI ops is a real discipline with real complexity. That is a career.
Data engineering. The quality of AI output is bounded by the quality of data input. Getting enterprise data into a shape where AI can use it effectively is genuinely hard work. Retrieval pipelines, embedding strategies, chunking approaches, metadata enrichment. That is a career.
Prompt Skill Is Table Stakes
I am not saying prompt skills do not matter. They do. Every person working with AI should understand how to write effective prompts. But that is precisely the point. It is a skill everyone needs, not a role one person fills.
It is like saying "typing" was a job in the 1990s. Sure, some people typed faster. But you would not build a career around it because eventually everyone had to type. Prompting is the same. Within two years, writing effective prompts will be as basic as writing a clear email.
The people I see building actual careers in AI are the ones who treat prompting as one tool among many. They understand model architectures. They can evaluate outputs systematically. They can design systems, not just write inputs.
What to Actually Hire For
If you are building an AI team, here is what to look for:
AI engineers who can build end-to-end pipelines, handle model integration, and write production-grade code that wraps AI capabilities in reliable systems.
Evaluation specialists who can design test frameworks, define meaningful metrics, and build automated pipelines for continuous AI quality assessment.
AI architects who can design the system-level decisions: model selection, routing strategies, data pipelines, integration patterns, cost management.
Domain experts with AI fluency who understand your industry deeply and can identify where AI creates genuine value versus where it is theatre.
Notice none of those are "prompt engineer." Because prompting is something all of them do as part of their actual work.
The Uncomfortable Truth
The prompt engineering hype served a purpose. It got people interested in AI, gave them a low-barrier entry point, and created a community of practitioners who could share knowledge. That was genuinely valuable.
But we need to move past it. The companies that are still hiring dedicated prompt engineers in 2025 are optimising for the wrong thing. They are hiring someone to talk to the model when they should be hiring people to build, evaluate, and operate the systems around it.
The prompt is five percent of the problem. The other ninety-five percent is where the actual work lives.