I saw a guy on LinkedIn yesterday with ‘Senior Prompt Engineer’ in his headline. He had 12,000 followers and was selling a $97 PDF of ‘secret hacks’ to make Claude 3 Opus sound more human. I almost threw my laptop out the window. It’s the biggest grift in tech right now, and I’m tired of pretending it’s a real job. It isn’t. It’s a temporary workaround for the fact that AI is currently a bit clunky. By this time next year, that title is going to look as ridiculous as ‘Master Horse-Whistler’ looked once the Ford Model T started rolling off the line.
Prompt engineering is a bug, not a feature. We only need it because the models aren’t smart enough yet to understand what we actually want. But they’re getting there. Fast.
The part where I admit I fell for it
I wasn’t always this cynical. Back in early 2023, I was obsessed. I spent about three weeks in a dark room trying to figure out the exact sequence of tokens that would make Midjourney v5 stop giving people six fingers. I even bought one of those stupid prompt packs for $25 from some guy on Twitter. I thought I was ‘upskilling.’ I thought I was becoming a wizard who could talk to the machine. What a joke. I spent four hours on a Tuesday afternoon—I remember it was raining and I was at that overpriced Blue Bottle on 2nd Street—trying to generate a specific image of a brutalist library in the snow. I used words like ‘hyper-realistic,’ ‘8k,’ ‘octane render,’ and ‘volumetric lighting.’ Every result looked like a muddy mess or a video game from 2012.
Then my niece, who is seven and doesn’t know what a ‘token’ is, sat down and typed: ‘A big concrete building with books and it’s cold outside.’ The result was perfect. Better than anything I’d spent all morning ‘engineering.’ That was the moment I realized the whole industry was built on sand. The better the AI gets, the less your ‘engineering’ skills matter. If the machine can understand a seven-year-old, it doesn’t need a specialist. It just needs a person with an idea.
Anyway, I think I’m still bitter about that $25. But I digress.
Why the tech is making you irrelevant
The whole premise of prompt engineering is based on the idea that LLMs are static. They aren’t. Every update to GPT-4 or Claude or Gemini is specifically designed to bridge the gap between ‘what you said’ and ‘what you meant.’ We are moving toward a world of intent-based computing. You won’t need to tell an AI to ‘act as a world-class copywriter with 20 years of experience in the SaaS industry.’ That’s just filler. It’s fluff. The model will already know the context of your business, your tone of voice, and your goals because it will be integrated into your entire workflow.
I tracked my own usage over the last six months. In October, I was writing prompts that were 300 words long, filled with constraints and examples. By March, my average prompt length dropped to about 15 words. Why? Because the models got better at inferring. I did a test: I ran 50 identical tasks through GPT-4 in November and then again in April. The April results were 30% more accurate despite using 50% less instruction. The skill isn’t in the syntax anymore. It’s in the vision.
If your only value is knowing how to talk to a computer, you’re in trouble. The computer is learning how to talk to everyone.
I know people will disagree with this. They’ll say that ‘complex chains of thought’ and ‘multi-step reasoning’ require a specialist. I think that’s cope. It’s people trying to protect their niche. Companies like Jasper or Copy.ai—which I personally find almost unusable now because they’re just wrappers for things I can do myself for free—are scrambling to find a new angle because ‘prompting’ is no longer a moat. It’s a commodity.
The ‘English Major’ revenge (that won’t actually happen)
There’s this popular take going around that ‘English majors are the new computer scientists’ because they know how to use words. I used to think this too. I was completely wrong. Being good at words doesn’t make you a good prompter, because eventually, words won’t be the primary interface anyway. We’re going to be using sketches, voice, gaze tracking, and existing data sets. What I mean is—actually, let me put it differently. The interface is disappearing. If you have to spend ten minutes ‘writing’ to get an output, the technology has failed.
I have a friend who works at a big agency in Chicago. They hired two ‘Prompt Engineers’ last summer. By January, they had transitioned them into ‘Creative Strategists’ because they realized that ‘engineering’ a prompt takes about five seconds of trial and error for anyone with a brain. The real work was figuring out if the output was actually any good. That’s the piece that people are missing. Taste is the only thing that isn’t going to be automated.
I might be wrong about this, but I actually think we should stop teaching ‘AI literacy’ in schools and just go back to teaching art history and philosophy. If the AI can do the execution, the only thing left for the human is the judgment. And you can’t ‘prompt’ your way into having good taste. You either have a perspective or you don’t. Most of the ‘prompt engineers’ I see online have the aesthetic sensibility of a corporate PowerPoint template. They’re obsessed with the ‘how’ because they have nothing to say about the ‘why.’
Specifics matter, but not the way you think
Let’s look at some actual numbers. I ran a small experiment with my team last month. We had two groups. Group A was ‘trained’ in prompt engineering techniques (Few-shot, Chain of Thought, etc.). Group B was just told to ‘get the best result possible’ using their natural language. We tracked 20 different creative tasks over two weeks.
- Group A spent an average of 12 minutes per prompt.
- Group B spent an average of 2 minutes per prompt.
- The ‘quality’ rating—judged blindly by a third party—was 4.2/5 for Group A and 4.1/5 for Group B.
The ‘engineers’ were 600% slower for a 2% gain in quality. That is a terrible ROI. It’s a waste of human life. Total lie.
I also have a weird, maybe unfair, hatred for the term ‘Engineer’ being used here. An engineer builds bridges. An engineer writes code that handles edge cases and manages memory. Typing ‘/imagine: cat with hat’ into a Discord bot is not engineering. It’s barely even ‘typing.’ Calling it engineering is an insult to people who actually build things. It’s like calling someone who uses a microwave a ‘Molecular Gastronomist.’ It’s pretentious and it’s a red flag for any hiring manager who actually knows what they’re doing.
I refuse to use any tool that markets itself primarily on ‘prompt libraries.’ I’m looking at you, various Chrome extensions that I won’t name but we all know the ones. If I need a library of 10,000 prompts to use your product, your product is broken. I’ve uninstalled three of those in the last month because they just clutter the UI and offer nothing but clichés.
What happens next?
So, if prompt engineering is dead by 2025, what’s left? I think we’re going to see a massive return to ‘craft.’ Not in the sense of doing everything by hand, but in the sense of having a deep, visceral understanding of the medium you’re working in. If you’re a designer, you need to know why a certain typeface works, not just how to ask an AI for ‘cool fonts.’ If you’re a writer, you need to know how to structure a narrative, not just how to ask an AI to ‘write a blog post about AI vs Human Creativity.’
The AI is a tool, like a hammer. We don’t have ‘Hammer Engineers.’ We have carpenters. The focus is shifting back to the output, not the mechanism. I’m honestly relieved. I’m tired of talking about prompts. I’m tired of reading ‘Top 10 Prompts to Explode Your Productivity’ threads on X. It’s all noise.
I don’t know exactly what the job titles will be in 2026. Maybe we just go back to calling people ‘Writers’ and ‘Artists’ and ‘Analysts.’ That would be nice. I’m looking forward to a world where the technology is so good that it becomes invisible, and we can finally stop talking about the plumbing and start talking about what’s actually coming out of the tap.
Will I be out of a job? Maybe. But at least I won’t have ‘Prompt Engineer’ on my resume when the bubble pops.
What are you going to do when the machine finally understands exactly what you mean the first time you say it? That’s the question we should be asking.
Stop buying the PDFs. Just talk to the damn thing.