my take on ai
ai is everywhere. in your car, your telly, almost every conversation you have and maybe even in your pet. More, ai can and will do everything, anything. Save the world, end mankind, probably both at the same time. Kill your job, your pet, or maybe you get killed by your pet in your sleep, or while you are watching Terminator the 12th time.
So, what's not to hate about ai? Or, if you are an optimist, not to love?
For me, as a software engineer and developer, the case is was clear. When ai arrived in our world, only the dumb jobs where killed. You could speak or at least translate any language to any other language (heck, even elvish was on the menu), let the ai write your follow up mails on a topic nobody really was interested in, skip google search for most things you wanted to know. A glorified search bar that i could talk to, that understood my mumblings and produced text that felt like there was a spark in there. Something like, if not intelligence, then at least understanding and reasoning.
then it became better. fast.
All of a sudden, it wasn't the obvious jobs ai took over. Picture generation, movie clips, music, text. Creative work. I got annoyed, because honestly, those pieces of art where anything but that. Until now, there was not one thing ai could possibly invent we - us humans - didn't think of first. There i was, annoyed but again, not feeling threatened by the overmind.
then there was coding agents
All of a sudden, almost overnight, maybe actually from sunday night to monday morning, the ai could and would produce code. Write thousands of lines of code, in any programming language, at a speed 2 million monkeys on typewriters could not match even if they got the golden banana. And it produced code that seemed like it was written by somebody with skill.
At this time, I was still playing with half-brained code completion tools in my favorite ide, laughing about the robot vacuum cleaner losing its head over where it had already cleaned and not finding the docking station. Fixing mistakes my ide integrated code helper made. Completely ignoring what was happening, the change that was going to come. Still believing it took a human brain to make software work.
Trying to figure out how i could use the ai to speed up my work, to make my day as a professional coder and software engineer easier.
And not wanting to throttle the next ai evangelist with a chatgpt cheat sheet.
Now the ai wanted my job
Of course it didn't want it. The ai is incapable of wanting.
This might be one of the most important things we have to understand when dealing with it.
The journey from ignorance to curiosity
Well then. It seemed like i couldn't sit out that one.
If you can't kill it, love it. Or at least use it.
It has a certain appeal, to be honest. Tell the thing what you want, go have a coffee or five, have a nap, and when you come back, it's done. Just like that.
Reality begs to differ, of course. And the reasons why i didn't get what i wanted where plentiful, and in some cases, even amusing.
After playing with chatgpt for some time, with very mixed results, prompt optimization was coming up.
Of course, to get what i wanted, i did have to tell the ai exactly what i wanted. So i worked on my prompting skills, to some but finally little avail. The results did get marginally better, but at the same time, more frustrating. That damned thing kept forgetting what we accomplished just 2 chat windows before. Invented conversations we never had. Took clues from wherever they did roam. Built something truly astonishing over the course of 2 hours, then messed it up utterly in 2 minutes, while "thinking".
Since this was not helping my daily coding business, I went back to old school coding, more than a little disappointed. Now even my ai powered ide auto complete did not satisfy. I've seen what ai could be, a truly helpful coding assistant. But it had the brains of a 2 year old, and all of the attitude.
Some weeks later, prompt engineering was on. Same as before, but with more context. Meaning, you could throw part of the answer to the ai and hope it would find whatever you wanted there. Context windows got larger, you could add something like a new york city telephone book to your prompt and then tell the ai to find all persons named Edgar having a phone number with at least three times 6 in it. And it would.
Progress, in a way
Then, there came coding assistants with mcp servers and agentized workflows. Fancy words for they can stick to whatever they are doing and actually do it. And what it did! At that time, vibe coding was a thing only in the minds of some people who got rich on ai hypes. Now, what it means is "you tell the ai what you want and it makes it happen".
For the first time, you could tell the coding agent what you wanted and, with minimal supervising, virtually no programming skills and a ai-account somewhere, that agent not only produced code but it produced code that worked!
If it was that simple, this story would have never got written, because by now i would not want to spend the rest of my useless career in front of a computer
The challenge of speaking to a very capable, very stupid genius that knows everything and nothing at the same time
As i said, you get what you tell the ai you want. There is absolutely no room for you know what i mean. No room for i told you so 3 times already. No room for if you do not do what i say this very minute...
All the mistakes are yours to make. All mistakes are yours.
If something does not work the way you wanted - your fault. Look what you asked for.
Not precise enough? Too much room for interpretation (never do that with an ai coding agent, by the way)? Didn't think it through? Maybe…do you really know what you want?
This, maybe more than anything, gave me a glimpse of how my field of work will change. And most probably, not only my field of work
With the ai, you've got a very fast, capable assistant at your hands. One that has access to much more information and data you will ever have, or will ever be able to handle. An assistant that can process tons and tons of data, in virtually no time, multithreaded, tirelessly. One that never gets sick, bored, fed up or angry. One that does exactly what you tell it you want, and that's the catch - the ai doesn't have to learn what you want.
you have to learn how to speak and utilize the ai
That means overcoming frustration with your own disability to make yourself understood. Frustration about how you did not think your way through until the end. Working, learning at the very edge of your ability to interact with something that maybe feel intelligent, or maybe even conscious, but isn't.
And after that, you will have to doublecheck whatever came out of that prompt window. Again, again, and again again.
At least, for now.
Conclusion - if any
The ai is a tool. Maybe the most mighty tool mankind ever made. Maybe the most dangerous, maybe the most brilliant. Maybe just another tool like wheel, book, internet, automatic cat food dispenser. Time will tell.
But what it isn't - a solution for all and everything.
You still need a problem to solve to use ai to solve it
Will it kill jobs, make people wish it was never intended? To be sure.
Can it help you run your company more efficiently? Most probably.
Can it help you make more money? Maybe.
Do you want how to utilize this tool?
Explore the opportunities, learn about the pros and cons and, most important, the dangers relying on ai as a tool?