> I think what’s interesting about AI, and why there’s so much conversation, is that in order to be a good user of AI, you have to really understand software development.
This I agree with completely. You can see it in the difference between a prompt where you know exactly what you want and when things are a little woolley. A tool in the hands of a well trained craftsperson is always better used.
> So I think we’re going to keep talking for quite a while Me neither, and to be clear I'm okay with that. This was mostly a rant at the lack of diversity of discourse.
Agree, the diversity of the discourse is not great. There's a lot of "omg I just got started waaauw" articles out there along with "we're all gonna die!" stuff. And then a few seams of very excellent insight.
Deep research at least helps with dowsing for the knowledge...
{heart}
https://www.youtube.com/watch?v=Q0iuMByFjCQ&list=PLEouLkiLHd...
;)
This isn’t to say there’s not hype. Just that if you’re not seeing big productivity gains you need to make sure you really are an outlier and not just surplus to requirements.
Rather, I hear a lot of nuanced opinions of how the tech is useful in some scenarios, but that the net benefit is not clear. I.e. the tech has many drawbacks that make it require a lot of effort to extract actual value from. This is an opinion I personally share.
In most cases, those "big productivity gains" are vastly blown out of proportion. In the context of software development specifically, sure, you can now generate thousands of lines of code in an instant, but writing code was never the bottleneck. It was always the effort to carefully design and implement correct solutions to real-world problems. These new tools can approximate this to an extent, when given relevant context and expert guidance, but the output is always unreliable, and very difficult to verify.
So anyone who claims "big productivity gains" is likely not bothering to verify the output, which in most cases will eventually come back to haunt them and/or anyone who depends on their work. And this should concern everyone.
The only downside is not learning about method Y or Z that work differently than X but would also be sufficient, and you don’t learn the nuances and details of the problem space for X, Y, and Z.
I've learned about outbox pattern, eventual consistency, CAP theorem, etc. It's been fun. But if I didn't ask the LLM to help me understand it would have just went with option A without me understanding why.
Software engineers who haven’t tried these tools don’t understand what they are, and vibe coders who never understood software are taking the mindshare in public because it sounds revolutionary to some and apocalyptic to others. You have to stop listening to the claw bros and try using these as tools yourself in small ways to see what it’s really about, IMO.
"Just verify" is glossing over a lot of difficult work, though. It doesn't just require checking whether the program compiles and does what you wanted—that's the easy part. Having the tool also generate tests and documentation is something you should manually verify as well. And at each step of this process there will likely always be something you missed, which considering you're not experienced in X, Y, or Z, you have no experience to properly judge.
You can ignore all of this, of course, which majority of people do, but then don't be surprised when it blows up in your face eventually.
And verification is actually relatively simple for programming languages and software. In many other fields and industries verification is very impractical and resource intensive. It doesn't take a genius to deduce the consequences of all of this. Hence, the net effect of these tools is arguably not positive.
I run static analysis on mixed human/AI codebases. The AI parts pass tests fine but they'll have stuff any SAST tool flags on first run — hardcoded creds, wildcard CORS, string-built SQL. Works in a demo, turns into a CVE in prod.
And nobody's review capacity scaled with generation speed. Most teams don't even have semgrep in CI. So you get unreviewed code just sitting in production.
The "10x" is real if you count lines shipped. Nobody counts the fix cost downstream though.
No, its verify that X approach is semantically correct, architecturally makes sense, design is valid and then add tests and documentation. Basically, 80% of the work.
I think people here are thinking I’m building Gmail from scratch when I’m talking about adding additional database APIs and models in the style of stuff I’ve already done twenty times in the same application. That’s easy for even the dumbest LLM and verification is not more than a code review, maybe an hour or two of labor. It’s never perfect, but as stated elsewhere I know how to program this stuff already so I can just fix it on the spot before I commit it.
I understand though, some of you either can’t use agents to code because of one reason or another, or refuse to; both are valid. I’m saying that for my job programming in a specific industry for a specific application in a specific language that AI agents actually help me out more than hinder me, and I still ship quality code in the style that matches the code base.
This is overly dismissive, there are many things that are possible now that weren't before because writing the code is no longer the bottleneck, like porting parts of the codebase from managed to unmanaged for teams with limited capacity. Writing code is about 1/3rd of the job. Another 1/3rd is analysis, which also benefits from AI allowing people who aren't very good at it to outperform. The final 1/3rd is-
> the effort to carefully design and implement correct solutions to real-world problems.
That's problem-solving - that part doesn't get sped up, and likely never will, reliably.
I don't mean to snipe at AI, because it really does seem to have set more people on the path of learning, but I was writing VB5 apps when I was 14 by copying poorly understood bits and pieces from books. Now people are doing basically the same but with less typing and everyone thinks it's a revolution.
There is more to it than "being able to make an entire application", which a novice could also have pulled off in a weekend 10 years ago.
You might not consider this productive, but they do, so what you think literally doesn't matter to them.
I don't believe they said that folks new to AI can't make impressive use of it. They did however say that senior folks with lots of scrappy and holistic knowledge can do amazing things with it. Both can be true.
They still have absolutely no clue how it works, so how could they "write entire applications"? They vibed it, but they certainly didn't write any of it, not one bit of it, and they're clueless as to how to extend it, upgrade it, and maintain it so that the AI doesn't make it a bloated monstrosity of AI patches and fixes and workarounds that they simply could never begin to understand.
They were also following a dozen youtube tutorials step by step, so even that part was someone else doing the thinking.
Yeah, these are the same guys constantly bugging me to help them figure something out.
I’m not a naysayer by any means and at this point use Llms all day for many purpose. But it is undeniable that the exact moment complexity reaches a certain threshold, LLMs need more and more guidance and specific implementation details to make worthwhile improvements. This same complexity threshold is where real world functionality lives.
I think it is genuinely impressive to be able to build one app with AI. But I haven't seen evidence that someone could build a maintanable, scalable app with ai. In fact, anecdotally, a friend of mine who runs an agency had a client who vibe coded their app, figured out that they couldn't maintain it, and had him rewrite the app in a way that could be maintained long term.
Again, I'm not an Ai detractor. I use it every day at work. But I've needed to harden my app and rules and such, such that the AI cannot make mistakes when I or another engineer is vibing a new feature. It's a work in progress.
Well, there was also a lot of unrelated things that happened as well around last November for me, but yes, getting into vibecoding for real was one of them, and man I feel physically drained coming back from work and going to use more AI.
Not sure what it is. I'm using AI personally to learn and bootstrap a lot of domain knowledge I never would have learned otherwise (even got into philosophy!, but man is it exhausting keeping up with AI. I would burn through a week's worth of credits in a day, and now I haven't vibe coded a week.
I think, I will chill. One day at a time.
My take is that it's similar to what Amber Case described in Calm Technology - with AI you are not steering one car, you're really steering three cars at the same time. The human mind isn't really designed for that.
I am finding that really structuring my time helps in terms of fighting back. And adopting an hours restriction, even if I could rage for 4 more hours, I don't. Instead I stop and go outside.
AI makes a ton of bad decisions too and it's up to you to work with it. If I had the knowledge of the dangers hidden in things I'm developing, I'd move even faster
Was able to make a great full web app, which I think is hardened for prod but it had to be refactored to do so. Which it happily did.
It's really about asking the right questions, breaking down tasks, and planning now. I'm going to tackle a huge project, hoping to share it here.
Me too. A key purpose of HN, and a bright time for that.
Now I'm working on a second project, all with AI. I haven't written a single line. It works better than a non programmer would make because I knew what to ask for. But I'll admit I'm not learning anything.
https://github.com/hparadiz/evemon
The core if you disable all the plugins is currently topped out at 73.8 MB after several days of running it. I've given it several audits with the AI agents using actual memory maps and doing the math on each variable.
I haven't had time to do Milkdrop yet but it's on my todo. The issue isn't doing the work. The issue is not having enough credits in my accounts to throw some compute at it. I'll get to it eventually. But it's actually way easier now to try new ways of packing the data into binary and profiling it for issues.
The issues I've had are edge cases like a 6 hour youtube stream. At one point the BPM detector was buffering the entire track in the pipewire sink. It took one throwaway prompt to the AI to solve that one.
Then the same happened with languages that managed memory.
And with IDE that could refactor your code in a click and autocomplete API calls.
And with Stack Overflow where people copy/pasted code they didn't understand.
But with that said, those who learn the underlying mechanisms will always be able to solve more problems than the folks who don't. When you know the lower pieces, your mental model tells you when and where the higher level pieces are likely to break. Legit superpower.
I would define that as being "seriously hamstrung"
So yeah i mean - who cares how it works - but also if you have experience in how things _do_ work you can solve problems other people cannot.
Yet most programmers nowadays can't write ASM or C and still manage to produce useful software.
For example, I haven’t racked and cabled a server in over 15 years. That used to be a valuable skill.
I also used to know how to operate Cisco switches and routers (on the original IOS!). I haven't thought about CIDR and the difference between a /24 and a /30 since the year 2008. A class IP addresses, how do those work? What subnet am I on? Is thing running on a different VLAN? Irrelevant to me these days. Some people still know it! But not as many as in the past.
The late Dr. Richard Hamming observed that once a upon a time, "a good man knew how to implement square root in machine code." If you didn't know how to do that, you weren't legit. These days nobody would make such a claim.
So some skills fade and others rise. And also, software has moved in predictable cycles for many decades at this point. We are still a very young field but we do have some history at this point.
So things will remain the same the more they change on that front.
And there'll be a split too... like there's a giant divide between those mechanics who used to work on carburetors and the new gen with microcontrollers, injection systems, etc. People who think cars are 'too complicated' aren't wrong, but for someone who grew up in the injected era, i vastly prefer debugging issues over the canbus rather than snaking my ass around a hot exhaust to check something.
Also anyone making a homelab has to know these stuff.
I don’t know if it’s the Universe delivering this farce or it’s the emergent LLM Singularity.
That's not what I said. I said that those who are already shining, are now shining even brighter. Give a great craftsman a new tool and he will find a way to apply it. If it is valueless, he will throw it away.
For what it's worth, your comment is also an HN trope, the disaffected low-effort armchair keyboard warrior.
It's what happened with the internet and computer usage. As Apple made it easier to get online with zero computer knowledge, suddenly we're electing people like donald trump.
Turns out Lowtax was right and ahead of his time
AI is the thing that for the first time can think better than us (or so at least some people believe) and is seen as an efficiency booster in the world of cognition and ideas. I'd think Hannah Arendt would be worried with what we are currently seeing and where we might be headed.
We have hundreds and thousands of years of history showing humans committing atrocities against each other well before the advent of computers, or even the introduction of electricity. So while the tool may become so ubiquitous that there’s no option not to engage with it, I don’t think it really fundamentally alters the dynamics of human behavior.
Some people are motivated by greed. Others are motivated by nobility. It really just comes down to which wolf they're feeding.
In terms of the tool keeping people ignorant, there’s a part I agree with and a part that I don’t. I think, in terms of information dissemination, AI is probably the autocrat’s wet dream in terms of finally being able to achieve real-time redefinition of reality. That’s pretty scary, and I’m not sure what to do about it.
On the other hand, people have always been free to not really learn their craft and to just sort of get by and make a living. That was true a thousand years ago, and it’s true today. There’s always somebody who can do really a high-quality job, but they’re very expensive, and then there's a vast population who will do a medium to terrible job for less money. You get what you pay for. There's a reason history is primarily written about people with power and wealth, they were the only ones with the means to do anything.
I don’t agree with the assertion about the internet and the election of someone like Donald Trump. Well before the internet existed, politicians were using communication mediums to influence things and get elected—whether it was the telegraph, the telephone, or the TV. JFK famously was the first TV president (notably, he didn't wear a hat).
These technologies simply give politicians more reach, and they may change the dynamics of how voters are persuaded. But what’s true today was true three hundred years ago: there’s the face of power that you see publicly, and then there’s what really happens behind the scenes.
Spoken like someone who thinks they are going to be insulated from the fallout
Sure, it might hurt me personally. I'm not selfish enough to put that over what will be an incredibly empowering development for our species.
This will be good for a handful of elites and no one else
Where are they flying and why software has gone to shit?
Maybe this super stars programmers have to keep their reality breaking technology secret, but everything has not only degraded, but turned to absolute trash.
But that said, this will be the 3rd major industry transition of my career. And having survived the past two, you will adapt, or you won't have a job. And that's why, once again, you will ultimately adapt, kicking and screaming if that's what it takes, so why not start early?
AI coding agents are useful already, but they make too many mistakes and they need handholding from expert engineering talent in 2026. Ask me again in 2027. But that's why the best results are coming from the talent right now with the experience to ask the right questions and propose the right tests and fixes as the human in the loop. Otherwise, it's still hallucinatory vibe coding in a loop IMO.
The surprise and disappointment, well not the surprise, is the usual hatred of success that defines humanity. Whatever, downvotes, right?
Yup
AI seems great when you have no way of truly validating its output.
It's really hard to separate the wheat from the chaff at this point but I've been positively surprised by the relatively few articles sharing their more advanced workflow, lessons learnt which helps me to avoid the traps, patterns emerging that taught me something new (or at least validated approaches I tried on my own which worked). Gets tiresome to keep pace so I try to not fall for FOMO, and avoid experimenting too much to not get lost until I see a pattern emerging from different sources.
I am amazed at the incredible things it can do - only to turnaround and not be able to do a simple task a child can do. Just like people.
It's also a nasty tool used to be dismissive of criticism by tearing people down.
Made even nastier by the way the character assassination can be done in such work-friendly language.
It's not a revelation to say profession has turds that shouldn't have jobs. Mostly working for bosses that need warm bodies and/or are themselves turds unable to really evaluate their performance. We been knowing this. It became impossible to ignore in the circa 2010 push to expand the hiring pool and lots of people that didn't even pretend to care about tech showed up.
It's just super convenient that we're suddenly calling them out when there's a need to dismiss informed criticism of AI.
But at the same time the more I read about AI, the more I realize I need to learn about AI. Thus far I'm just using cursor and the Claude code extension alongside obra superpowers, and I've been quite happy with it. But on Twitter I see people with multiple instances of Claude code or open claw talking to each other and I don't even know how to begin to understand what's going on there. But I'm not letting myself get distracted — Claude code and open claw are tools. They could go away at any time. But systems thinking is something that won't go away. At least, that's my gambit.
Does it write good code? I dunno. But it looks cool, and I think interesting in its own right, even if it ends up being functionally useless.
Since COVID I have seen teams scaled down, lots of custom development or devops/infra work work got replaced with SaaS and iPaaS cloud products, serverless/lambda, managed containers.
This is the next step.
Great that people feel more productive, unfortunely for many of them, us, more productivity means the C-suites can do some head count reduction yet again.
Specialists/generalists, top-down/bottom-up, BFS/DFS, pragmatists/idealists, ADHD/ASD; lots of continuums in software work and those at either extreme have biases.
Personally I think that there will be less programmers needed and the ones that will remain will have had to mellow out towards the center in all these continuums. We won't be able to rely on big teams balancing each other out.
Generalists will need to learn which details matter. Specialists will need to learn the delegation and risk tolerance usually reserved for the bemoaned management track. Hard to say which is the easier journey.
I think what’s interesting about AI, and why there’s so much conversation, is that in order to be a good user of AI, you have to really understand software development. All the people I work with who are getting the most value out of using AI to deliver software are people who are already very high-skilled engineers, and the more years of real experience they have, the better.
I know some guys who were road warriors for many years —- everything from racking and cabling servers, setting up infrastructure, and getting huge cloud deployments going all the way to embedded software, video game backends, etc. These guys were already really good at automation, seeing the whole life cycle of software, and understanding all the pressure points. For them, AI is the ultimate power tool. They’re just flying with it right now. (All of them also are aware that the AI vampire is very real.)
There’s still a lot to learn, and the tools are still very, very early on, but the value is clear.
I think for quite a few people, engaging with AI is maybe the first time ever in their entire career they are having to engage with systems thinking in a very concrete and directed way. Consequently, this is why so many software engineers are having an identity crisis: they’ve spent most of their career focusing on one very small section of the overall SDLC, meanwhile believing that was mostly all there was that they needed to know.
So I think we’re going to keep talking for quite a while, and the conversation will continue to be very unevenly distributed. Paradoxically, I’m not bored of it, because I’m learning so much listening to intelligent people share their learnings.