“Back then” vs “Now”
48yo. I don’t even remember my whole experience. The most accurate record of my experience is whatever I managed to document on LinkedIn.
Did I start coding at 3, 4, or 5yo ? Honestly, who cares? I certainly don't. It has been long enough that the minor details (Basic, asm 68k, perl, …) just don't matter. The important bit is that I don’t even want to write code anymore. Not that I really ever enjoyed coding anyway, with some minor exceptions. I simply wrote what I had to for my projects, ideas, or because the job required it.
Now I just let Claude Code do its magic under my direction. And that… I like it. A lot.
Whenever I have an idea (and I have A LOT of ideas) I just let Claude write it. It works. The idea is either worth exploring, or trashed 5mn later for some reasons: not what i expected, not practical, not enough compute (I never have enough compute), it doesn’t matter that much. It’s a 5 to 15mn job for Claude. If I spend more time on it then it passes the “worth exploring” checkpoint.
Ideas aren’t worth anything, only the implementation counts.
I’ve heard that for so long… It was a lie back then, it’s totally wrong now. Code doesn’t matter much anymore in most cases.
It still matters in some critical code, field specific areas (Medical, Embedded, Real Time, anything very hard technical & governance constraints). But for how long ?
Now the entry barrier to code is so low that anything goes:
- Everyone talks about AI safety for some reason. But why ? This self-censorship is frustrating, isn’t it ? Well… Let’s jailbreak one (done) and see what happens… Oh no… No-no-no-no ! Oof ! Okay, I get it. It’s critically important.
- What if I made an AI systematically, confidently, wrong ? (good idea, soon)
- “Sycophancy as a Service” (SaaS) sounds cool, let’s do that. done.
- Can I create a Reinforcement Leaning model to play a racing game ? Hey, let’s just code the racing game while I'm at it. Done and done.
- Can it play Checkers ? Sure
- Go 9x9 ? Woops… it works, but not enough compute to train it as a worthy opponent.
- Can I try to fine tune a non-reasoning model into a reasoning model ? I don’t know if it’s possible, but I failed to do so. Also, not enough compute.
- Can I teach math to a model ? Yes I can. But…. meh. Poor results. Need more compute, a DGX Spark specifically would be nice.
Hey ! What about the title ? clickbait ?
It’s engaging isn’t it ? Or at least it’s something I would give a try. It’s not clickbait though. It’s just that… I don't really have a good answer yet.
However,
Watching what people try to pull off with AI... it seems to matter. Why? I don't really know yet.
- Can I catch the obvious screw-ups early? Yeah, for sure.
- Better prompting? Please. "Prompt engineering" is just so 2025.
- Perhaps the fact that I have everything in a large mono repo ? It seems to help.
Every time I kick off a new project, I just let Claude scan the old work to pull out the "good bits." That architectural consistency across everything seems to work. Is it worth 40 years of experience ? Hell no. Or I hope not... It’s hardly a secret, but that’s an idea worth exploring.
Now I just tell Claude: "Hey, fix that thing about that bug. Just do whatever as long as it works". And it does.
Hardly a superpower, is it? Maybe my "superpower" is just that I'm not trying so hard to prove my worth by boarding every hype-train hoping one of them doesn't run out of steam.
Or perhaps, just perhaps, I didn’t develop any superpowers at all.
What now ?
I just created this account, on dev.to, to share my “worth exploring” experimentations, their potential results and failures. Not just about AI, but mostly AI.
And also something about me. As a coder, sysadmin, dba, hacker, tech lead, writer, devsecops, SRE, redteamer, race marshall, race director, biker, and some more.
Top comments (0)