AI is slowly munching away my passion
35 points by dzervas
35 points by dzervas
I used to be “the guy with that automates stuff”. I used to be the one to ask for “the company VPN hack that makes it work on Arch”. [...]
Now however, AI is the thing that automates and fixes corporate VPNs on outdated distros better than I would be able to in a thousand lifetimes. It has “robbed” me of a trait - or maybe a couple of them. [...]
Your trait isn’t taken away from you but it’s given to everyone, freely, with close to no required effort/skill/planning towards it.
I'm feeling this a bit around prototyping. For most of my career rapid prototyping was my super-power: I could knock out a working illustrative demo of an idea in an hour, then take that to a meeting and use it to elevate the conversation - my experience is that having a prototype to help solidify what's possible makes conversations so much more productive, especially if you know how to frame it so that it encourages rather than discourages new ideas.
Today, Claude can knock out better prototypes than I can in just a few minutes. My special trait suddenly isn't so special any more.
So far I've been handling this by leaning into it. I can produce a dozen prototypes in the time it used to take to create his one, which means I can explore a problem space faster and in more detail. That's an amplified version of my original trait.
It does feel weird to have a specific skill that I took pride in suddenly commoditized in this way, especially when I could name dozens more similar skills that are also affected like this.
My superpower was being able to write a minimal solves-my-problem script really fast. That often meant knowing a lot of weird languages that happened to be good at representing a problem, knowing how to quickly skim APIs, etc. AI is a 100x better than me now. It makes me more productive, but I still feel like I lost something special.
I was good at that too (though I wouldn't go as far as calling it my superpower). But to be honest, I also found writing such scripts a chore. Especially for things like data ingestion. I am pretty happy that LLMs can do this for me now, so that I can focus on more interesting problems. The interesting problems are typically more about thinking than writing code and if the domain is specialized enough LLMs (still) fail quite miserably.
That said, I'm pretty opinionated what Python, shell script, or Nix should look like, so I'm still often iterating on this kind of script with an LLM if it ends up being more than a one-off.
I'm in the same boat - "not knowing" something was never the problem, i could pick it up very fast to the extent of fixing the problem at hand
Wow, amazing skill.
I hope that you find your way through this and you're able to replace it with a better/faster/stronger trait. Maybe this will magically blow over and we'll get to how things were before - but I highly doubt that'll happen...
Mine was just raw attention to detail, which early in my career meant really thinking of everything. It's difficult to decide where to focus as things no longer fit into my attention and working memory. That was already a problem that grew with seniority and a skill to match, but now the problem side is quite accelerated and skill is playing catch-up.
Another case of Deep Blue.
(For added context - Deep Blue is the term we coined on the Oxide and Friends podcast for the sense of ennui that LLMs cause in many software developers - clip here: https://www.youtube.com/watch?v=lVDhQMiAbR8&t=2835s
EDIT: this inspired me to finally write about it properly: https://simonwillison.net/2026/Feb/15/deep-blue/ )
I’m glad you wrote it up. I was laughing like a crazy person when “Adam blew out the candles on Bryan’s cake”.
I remember how dismayed Kasparov was after losing the 1997 match to IBM's Deep Blue, although his views on Deep Blue became more balanced with time and he accepted that we had entered a new era in which computers would outperform grandmasters at chess.
Still, chess players can take comfort in the fact that chess is still played between humans. Players make their name and fame by beating other humans, because playing against computers is no longer interesting as a competition.
Many software developers would like to have similar comfort. But that comfort is harder to find, because unlike chess, building prototypes or PoCs is not seen as a sport or art-form. It is mostly seen as a utility. So while brain-coding a PoC may still be intellectually satisfying for the programmer, to most other people it only matters that the thing works. That means that programmers do not automatically get the same protected space chess players have, where the human activity itself remains valued even after machines become stronger. The activity programmers enjoy may continue but the recognition and economic value attached to it may shrink.
So I think the big adjustment software developers have to make is this: The craft will still exist and we will still enjoy doing it but the credit and value will increasingly go to those who define problems well, connect systems, make good product decisions and make technology useful in messy real-world situations. It has already been this way for a while and will only become more so as time goes by.
In my opinion, we are currently overestimating the impact of LLMs. While they are fantastic information retrieval tools, I remain skeptical that they can reliably ship software, regardless of how much reinforcement learning is bolted onto them.
Instead, we should leverage LLMs to automate mundane tasks while utilizing theorem proving, model checking, and other formal verification techniques to truly increase software quality taking advantage of the extra productivity gains.
I'm not arguing that LLMs do or will ship "production-grade" code (whatever that might mean). I'm expressing that the "information retrieval" part "stripped" me (or more accurately gave it freely to everyone) of my trait to be the first to ask when you want to do information retrieval or solve minor, very specific, annoying problems.
Weather LLMs are gonna take our jobs is another discussion - I don't think they will or I don't even care to tell you the truth. Let them, I'm gonna leave the computers behind and become a woodworker or something. The trait part though is something deeply personal that defines one in the context of their peers. It's not technical or professional, it's personal and even more, social
Fair enough, but I still think you probably undervalue your skills, and there are probably lots of really interesting high-value things you can do in CS & SWE where LLMs won't lower much the barrier of entry and/or won't compete with you.
I think the impact may be bigger long term by LLMs changing how software is "shipped", or what software even is.
When the cost to write code is high, we treat it as a precious artifact. We maintain it, update it, and ship the same code to all users. The quality must be high, because bugs can affect many users and long-term maintenance.
But I think there's a chance we'll get a lot more disposable single-use software. Programs created on the fly to solve one task and then discarded. Nothing to maintain. Whether they're buggy or not is judged only by an individual user being satisfied with results on a particular task.
I totally agree; to be honest, I also enjoy working on the higher, system-level they allow me to do so more and more.
Fundamental understanding of software languages, tools, libraries and architecture will continue to be an advantage, probably even more so then ever; especially as more devs will become resigned and lazy by default.
I'm still in a lot of ways the "tool guy" on my team, but it's in large part because I have good instincts about what tools will be useful to our development workflow. pretty much anyone on the team could write the tools, I'm just the person who thought of what tools would be nice to have. and even with claude in the loop I still need a fair amount of back and forth to get things into a good state.
the only winning move is not to play
What I'm really afraid of is that sw development is only the first frontier. I think a lot of us will fall, only a few will stand, but what's after us - everything else. How will they hold up? What happens after? My personal goal is to navigate to survive, as I still believe that problem solving and craft is the longest-term skills which can keep me above the water.
Same. I hope that something will pop in a special interest manner (like coding did) and slowly take its place.
For the rest I'm almost fascinated to see how it plays out - more of a "watch the world burn" though really
While vibe-coding a read-only, dependency-free front-end for raw Git repositories on self-hosted or shared-hosted servers (in PHP), adding syntax highlighting for new file formats in seconds feels incredibly rewarding and liberating.
I'd get no joy in writing those regular expressions. Providing the source files along with a prompt such as:
What are some other popular programming languages and source file formats that are missing? Add them.
And:
The Highlighter class identifies more syntaxes than are present in the LanguageDefinitions. Find the missing syntaxes and add them.
Along with cleaning up code:
Eliminate the early return (only one exit point per method). There's some code duplication going on inside the file. Can you spot it and eliminate it?
That is joyful. Seeing the code appear more quickly and more accurately than I could ever write. That is joyful. It'll be even more fun when these code monkeys can permanently learn aspects of professional-level code like, "don't violate encapsulation", "avoid insipid comments", "avoid duplicated logic," and so forth, without having to have a file full of rules.