AI "Prompt Engineering" is Not A Skill, It's Gambling
Programmers are taking the wrong approach to using AI!
To Prompt or not to Prompt?
The Magic of Vibe Coding
Vibe coding, prompting an AI agent without writing any code, can be a magical experience when you want code that’s similar to what LLMs were trained on.
AI excels at writing bash commands, scraping websites, and organizing data. This grunt work isn’t that fulfilling, so devs were stoked to offload it to AI.
But recently, since AI agents have greatly improved, some programmers got false hope they could do more novel coding too.
I used to enjoy programming, now I spend my days going back and forth with an LLM and pretty often yelling at it. Telling it that it’s doing the wrong thing and getting mad.
The problem is, it’s kind of impossible to identify the boundaries of what AI can and can’t do without testing it. Each time it messes up, my instinct as a curious, stubborn engineer is to believe that it’s my fault. So I rewrite the prompt and coax the AI to make corrections.
Sometimes re-prompting actually works when the AI just needs a bit more context.
But often I find myself getting caught in a re-prompting loop, like a gambler pulling the lever over and over, thinking that I’m so close to winning, when I should’ve quit while I was ahead!
One of the reasons I like programming is that it’s predictable, it’s logical, it’s knowable. […] And that works really well with my brain. […] But when we’re working with AI LLMs it’s not predictable. You can use the exact same prompt and get a different response every single time. […] That’s not what I signed up for.
No amount of prompt engineering guarantees the right answer.
Anyone can pull the lever on a slot machine, the same goes with prompting. LLMs were designed to be used through natural language, not incantations.
“Prompt engineering” courses are scams that prey on programmer’s bias towards misapplying logical reasoning and systems thinking.
Productivity FOMO
But now CEOs at companies like GitHub are demanding that developers embrace AI or leave the profession. This requirement does make sense, GenAI can sometimes do in mere minutes what’d take hours for a human developer.
But this productivity FOMO is pushing developers to overuse AI, with diminishing results.
Sure I used to get frustrated while programming before AI, but when I finally discovered how to fix a bug or implement a tough feature it felt rewarding! That’s cause programming is a skill, if you practice and learn enough, you’ll eventually figure things out.
Yet, since the novelty of vibe coding is wearing off, watching AI successfully complete tasks can be pretty boring. If I wanted to watch others do work I would’ve been a manager… ha!
Prompters Anonymous
But when AI fails and I go all-in arguing with it, I’m like a gambler going broke. I get even more frustrated thinking about all the time I wasted before realizing I need to start over and do the work manually. Failing at prompting is demoralizing.
It’s good to see programmers talking about the effects of AI coding on their mental health. If you’re having these issues, you’re not alone!
But seriously, the most important thing to learn about prompting is when NOT to. It’s time to start enjoying programming again. :)