This is true, but if you use a file system that does scrubs (and you should use such a file system for a NAS), and if you keep the NAS on 24*7, this won’t be an issue.
I’m always struck by a bit of wonder at comments like this. It seems everyone’s experience is all over the place. Curious, what types of things are you working on where you see these results?
I’m at 90%+ code AI generated by stats. I work in embedded systems. It still goes off the rails all of the time and needs a heavy hand to guide it. It does not currently feel like it will ever be truly able to operate independently. It’s a very useful tool, but it’s just not there yet in my day-to-day.
I'm in internal tooling for blue and network teams, 70% of my code is generated and I have the same experience as your. The two 'new' hires (1 year into the job, so not so new) however are probably in the same boat as GP, but they just don't see the work we have to put to rewrite their AI-generated issues into something legible and useful, or read their AI-generated PRs filled with duplicated/dead code.
Honestly, the biggest improvement AI have brought me is the complete end of my imposter syndrome. People say comparison is the thief of joy,but in my case, previously invisible devs getting more visibility, suddenly pushing new code every week at the same pace as me but clearly without any understanding any of the underlying issues did wonder to my self esteem. I still consider myself among the worst of my group, but my group evolved from 'all devs I know' to 'the competents', which is very nice.
Not sure how this works for y’all. On my team, they would get blocked to merge on code review until they met quality bar, and if this happened enough they would be having a performance issue that would be dealt with (they would be given multiple chances to course correct - polite talk from team lead, polite talk from manager (doc trail begin), not polite talk from manger, bad perf review / no raise + no RSU year, PiP, fired)
Well, I’ll start by saying look at what you’ve just said: you’re in embedded devices, one of the more niche software roles. 90% of your code is written by AI. You’re saying that you need to correct it sometimes. You’ve watched as AI goes from chatbots that literally cannot hold a conversation with a human being to LLMs that can do your job with supervision in less than five years.
Maybe your point is just what the difference is to me. It’s competence. On hard problems Fable can just iterate by itself like nothing before. You give it a task, it can plan it, formulate and falsify hypotheses, and do more complex reasoning than anything before it. The main difference is simply that it gets more right and can go deeper on everything. Yes, you do still need to course correct it, but it’s just on the next level.
It’s like what everyone experienced when Opus 4.5 came out last year, and the penny finally dropped that AI coding had arrived. If these sorts of jumps happen even one or two more times then it’s all over. It might even be over now and it’s just too expensive to run, but that will change over time.
If you can, you should try it and see for yourself.
Keep in mind that a good chunk of people here are basically AI boosters: they work for AI companies, sell AI courses and training and certifications, own shares or otherwise benefit from AI companies doing well. Oh, also, lots and lots of marketers. Gitlab kind of exposed this a while ago, their marketing division had alerts based on HN keywords, their marketers were literally expected to be on-call to these keywords and even had a sort of lite SLA to respond to those keywords (stuff like Github, they were supposed to insert themselves in the conversation).
To the direct point: I've used AI for a lot of things. It's good, you can definitely build some things faster. It's far from perfect.
Also, if we were sooooo close to AGI, why would Anthropic and OpenAI continue hiring SDEs like mad, which they are? Why would Claude & co have embarassing bugs years after release (the glitchy scrolling bug, their websites being fully synchronous in 2026!!)? Etc.
People just disagree, some people find that thought impossible unless their opponents are being paid, due to how obviously brilliant they are. And, that seems to be the minority position actually, mostly the discussion is the turkeys telling each other how Christmas is a myth.
The steam engine was invented in 1765 (+/-) but employment crashed in the textile industry starting in 1815, 50 years later.
LLMs were practically invented in 2017, 9 years ago.
Of course, software is software and we live in a faster age, but guess what, human needs and desires are infinite so requirements keep going up.
At this point I have 0 trust in purely agentic coding (i.e. the 0 supervision approach). YouTube randomly stops displaying subtitles, Gemini CLI (before its shutdown) and Claude Code suffer from constant refresh glitches, the Gemini and Claude mobile websites are synchronous, like we're living in the 60s (if I minimize Firefox both Claude and Gemini stop answering and Gemini until about 1 month ago would also lose my prompt - this is the kind of thing people were fired for in the past, etc). If this level of buginess is the best agentic coding can produce with unlimited state of the art mega corp tokens, software itself will collapse faster than the job market.
Claiming that there will be no more SDEs of any kind, worldwide, in max 2 years, is an extreme position.
I'm not even an SDE. If they're turkeys, AI boosters are pheasants :-p
And I know why the legitimate AI boosters do it. They were genuinely impressed by what they saw in agents, especially since many hated programming anyway. They extended the line into the future and assumed complexity was linear. These people looked at LLMs and agents and didn't see what was there, but what could be (frequently from a position of major SciFi consumers through their entire lives, as all techies are).
Optimism is a positive, extreme optimism is a great flaw.
> Claiming that there will be no more SDEs of any kind, worldwide, in max 2 years, is an extreme position.
Nobody said that. The claim is that within 24 months the models will have the capability required to replace all of us. I stand by that. How quickly that moves to everyone actually getting replaced is a social and economic question. There may very well be well be a long tail, I’m sure today you can find traditional weavers and barrel makers somewhere, they just happen to be economic novelties.
For the record, I’ll say that by 10 years out the profession is hollowed out enough for it to count as destroyed, and the process has already begun.
Ok... let's say the profession is destroyed. Who maintains all these ever increasing piles of code? Who figures out the requirements, who creates specifications? Who solves issues that are reported by users?
I like Amazon's service. Parking at Costco on a Saturday is absurd to the point there's memes about it. I really hate standing in lines. Delivery to my door is awesome and I'm willing to pay extra for it. I also see the Amazon truck going house-to-house and don't feel guilty: I'm just one more stop along the way, my marginal impact is nothing at this point.
I feel like it was the dot com boom. :s/"Make everything online"/"Make everything use AI"/g
In a similar story, AI makes sense for some stuff but not other stuff. The stuff where it does not make sense for is gonna do bad when the bubble pops.
The first trend of AI was "use as much as you can! You must use AI!!!". Hence the rise of tokenmaxxing leaderboards and KPIs on token use.
The second wave, happening right now, is "use it, but control cost". All the cool kid CEOs are now talking cost-control, rate limiting, and metering. If your management is a "follower" they are bound to hop on the trend.
LEO satellites are the size of a car and are spaced apart by the size of a state. They also all are in slowly decaying orbits and will fall out of the sky on their own accord in 10 years or less (they are designed with intentional structural weak points to break apart and burn up on entry). The concerns you have are valid and very real, and shared by the people designing these things.
That sucks. If GLP-1s work for you, more power to you.
Curious: how big of a calorie deficit did you run, and what was your macro (protein/fat/carbs) balance.
My personal experience is going low on carbs (especially added sugars) and high on fiber and protein made running a deficit suck much less in terms of feeling satisfied.
Also, a 10% deficit was okay (I was hungry but could mostly ignore it). A 25% deficit was very annoying and about as much as I'd care to do.
I did a lot of things, but all of them required a massive amount of sustained effort. I like that I can order the food I want to eat at a restaurant without looking up all the calories before I go. I like that I can just live my life without food thoughts taking up 60% of my brain space at any given time.
For some people, diet and lifestyle changes help control blood pressure. Other people need to take meds for the rest of their lives, anyway. We don't shame them for it.
reply