17 points by Odbyteusz about 16 hours ago | 19 comments | View on ycombinator
tripleee about 13 hours ago |
BuildPeter about 1 hour ago |
khaelenmore about 1 hour ago |
As "AI" slop eats up more and more of critical software and even tries to make it's way into hardware design, we'll start seeing more and more subtle nearly impossible to track errors: calculator gives 2+2 equals 5 why? Is it calculator? Underlying math lib? Compiler? OS kernel? RAM? CPU?
The thing is that we are too used to have some systems well engineered and rock-solid that we don't think about them. When javascript or python program fails, we don't blame the runtime, since we expect it to be stable. But can you expect vibe-coded slop to be stable?
And why would you even need a computer that cannot compute?
ak39 about 4 hours ago |
Short answer for anxiety: learn about the business your software skills serve and align with domain subject matter experts.
jimcollinswort1 about 13 hours ago |
In 5 years We will have far passed the ability and need to review code. We will have phenomenal AI reviews to do that.
Existing tools, systems, frameworks and such just won't matter, would just slow down AI. Replacing complete vertical system silos will be done in a single, crazy step, efficient to run, secured by an awesome security AI.
I recently retired after working in software development from 1975 to 2025 encompassing punch cards to AI.
muzani about 9 hours ago |
But I expect that SWEs will become more like civil engineers. Right now, the laborers are also the engineers. In the future, they'll be separated.
There's standards. We're just not following or aware of them because it was impractical. RFC 9078 on sending a thumbs up in an email. RFC 6531 lets you make an email address consisting of emojis. RFC 5322 on writing (nested) comments in the email address.
Then there's accessibility violations like re-entering passwords, lack of alternatives to drag and drop, toasts that aren't handled by screen readers, image carousels that can't be paused, relying only on red color to convey error.
We'll probably be communicating globally in a similar form to RFCs instead of open source code, just working out why/what/how and what goes wrong.
austin-cheney about 15 hours ago |
Silo 1 - This will be employers strongly forcing AI for most task completion. This will be most large corporate employers and most start ups. There will be little to no working knowledge of the internals of large applications and these employers will bleed money on outside consultants to occasionally provide manual maintenance and security hot fixes. Everything will continue to look good from the outside, but really its a messy house of cards.
Silo 2 - This will be institutional shops of skilled experts. Game companies that own internally developed rendering engines, kernel developers, independent open source, and systems architects come to mind. These will be people that might occasionally dabble in AI for various things, but they manually modify their code. These are experts who do not gain productivity by offloading their greatest challenges to AI.
Silo 3 - The consulting class. This is where the money will be. Technology design firms already do most of this, but the problems these consultants solve for their clients will be greater and the clients will be less capable. These people will explode in demand, but supply will likely remain unchanged, because experts are hard to grow and the corporate world won't produce them.
---
I also predict the job recruitment industry will struggle as demand for talent stratifies, because software as an industry struggles to identify talent. This problem will remain unsolved until something catastrophic happens and most employers need to simultaneously unload liability in ways they are currently forced to accept it internally. Fear of bankruptcy, loss of money, and complete destruction of business are not strong enough motivators to move the needle on this.
flincode about 13 hours ago |
But i was there when gpt2 and 3 came out i even had early access and I learned how it worked and I can say one thing companies still need software engineers as long as LLMs themselves exists, this isn't end all and be all they are trying to force this idea into existence AI by this time should have already replaced all coders but it isn't because giant corporations are trying to force scaling instead of finding a way to make them more intelligent in a meaningful way.
publlus_enigma about 11 hours ago |
A return to personal computing as it existed in the early 80s, where owning a computer meant writing your own software for it, with exceptions.
Alternatively, hardware scarcity meaning everything is forced to the cloud.
spottedmarley about 14 hours ago |
paulcole about 8 hours ago |
ethnrob about 16 hours ago |
if you're anxious about job security, standing out in your career is generally not specific to the industry you are in.
yuzzuwo about 1 hour ago |
mlacaze about 3 hours ago |
The onramp for coding becoming easier means non-tech people finally have a cheaper way to try out business ideas for domains they're experts in. Many of those will succeed and will need to hire developers, massively increasing the demand.
Coding becoming easier means many more people will jump in, reducing wages rapidly. While there might be some work available, it'll mostly be low paid.
Now that one person can do the job of 5, 4/5 will be laid off and we'll all struggle to get work.
Now that one person can do the job of 5, AND there will be an increased demand for more developers from more businesses popping up, we'll be paid a hell of a lot more for the value we can generate.
The speed of code generation is going to catch up to us in the form of tech debt, and we'll see the outcome of that in 12 months as codebase development slows to a halt and AI can't deal with it. We'll be swamped with work.
AI will continue to develop at a rapid pace and take over all white-collar work and the world will change as we know it.
Take your pick. I don't know.