6 ms·

AI Changes Everything

55 points by tosh 1y ago

Zoethink 1y ago
I genuinely feel like we’re living through another big shift — like what smartphones did in the 2000s, AI is doing now. I’m not a developer, but I can already feel how the way I use the internet is changing. I used to search for things, now I just ask — and the AI replies like a person. It’s subtle, but powerful. I really think in a few years, most of us will have some kind of personal AI. Not just a tool, but more like a life companion — helping us plan, decide, learn, even reflect. Kind of like how we can’t imagine life without Google now.
- inferiorhuman 1y ago
  Yeah it's definitely changed the internet. All of sudden every site is behind anti-bot garbage (usually Cloudflare). Sites I use (like Sourcehut) are getting DDOS'd by AI scrapers. It might be a wee bit reductive to say that AI is destroying the internet, but this certainly is the death of the open internet. AI has just handed a huge amount of power to companies like Cloudflare and made it that much more difficult to avoid being tracked.
csomar 1y ago
> Do I program any faster? Not really. This really got my attention. I use Claude heavily and although I am sure it is saving me considerable time, I feel like my "delivery" speed is roughly the same. I think the realization that will come from this is that writing code is not that consuming after all; unless you are doing cryptography or math. Most of your time is spent testing, deploying, figuring out some weird bug, writing docs somewhere, signing up to some service, etc...
- Frieren 1y ago
  While writing code I think about what I am doing, improvements, etc. To write easy boilerplate code helps me to find improvements or even realize that there is some risk I overlooked. It is a zen state that helps with the deep thinking part. Not writing code forces me to read and reread the code to get the same realization. That is why it is harder to work with code written by other people. In the age of AI all code is code written by someone else that I have to maintain, it looks like a nightmare.
  - fhd2 1y ago
    Yeah I have the same. In general, writing (and experimenting) is how I _think_. If I read code, it takes me a lot longer to understand. Sure, I save some time by not having to do any trial and error, I'm looking at a solution that already works according to some testing. But then I start to wonder about edge cases, leaky abstractions and such, and not having done the work, that's where a lot of the effort saved by not writing it comes back, at a stage where the work is seemingly already done and "just" needs reviewing, which is somehow more frustrating. Perhaps people are just different. I work great on a blank canvas. I know a lot of people struggle immensely with it. Hell, some people type so slow or have such low mastery of their tools, I really feel their pain watching them.
  - amelius 1y ago
    Yes, AI is taking all the fun jobs.
  - AlexeyBrin 1y ago
    One old trick that helps a bit with understanding, is to not copy paste what the AI suggested, but rather to manually copy the code suggested the way you would do it with a physical book.
- the_mitsuhiko 1y ago
  > Most of your time is spent testing, deploying I became hyper attentive to iteration speeds as a result. This is not just something a human runs into, the agent runs into it even more because it loves to run tests etc. The quicker that goes, the more efficient the agent is. Obviously the same should be true for humans but because the agent doesn't make breaks quite like a human does, it becomes more noticeable.
  - csomar 1y ago
    It gets me frustrated. Here you are one second flying, in the zone, moving from concept to concept at faster than human speeds; and then boof, you have to pull up some random api doc, open postman, test the api, come back to the chat console, etc...
    - the_mitsuhiko 1y ago
      What's nice is that a codebase optimized for agent also becomes a codebase optimized for humans.
      - saubeidl 1y ago
        Disagree. A codebase optimized for agents is overly verbose and takes more conscious effort to parse.
        the_mitsuhiko 1y ago
        Overly verbose codebases are not great for agents in my experience because they confuse them greatly and make refactoring hard. However codebases that I find to work really well with agents are quite lite on abstractions and I always felt that this is the right way to build software for humans too. I have a lot of functions, I pass a lot of data, I embrace a lot of thread local state and this works really well for both humans and AI to understand what is going on. I gave a talk on this recently (not the agentic view), but I found this to work exceptionally well for agents: https://www.youtube.com/watch?v=ej5RsTtVvQE https://www.youtube.com/watch?v=ej5RsTtVvQE But here is where agents and humans both agree: quick and easy tooling. Make it easy to iterate on a test, make it easy to manage resources, make the iteration cycle as quick as possible. I'm writing this as my partner's job forces iteration cycles of minutes on her. No agent can do better in that space than a human.
- herbst 1y ago
  I am undecided about AI coding. Ive done some vibe coding with working but code wise horribly results. But I have had working prototypes faster than ever. But I recently starting doing some redesign and architecture changes (manually) in an old code base and notice that i tab trough whole files auto fixing code one by one with suggestions which is amazing. I think we need to find a way to work with it and keep it far away from making architectural decisions for now.
- meindnoch 1y ago
  >writing code is not that consuming after all; unless you are doing cryptography or math Most cryptographic algorithms fit on a napkin. 99% of the work goes into proving the mathematical properties of the algorithm.
  - yencabulator 1y ago
    But that napkin may not be constant time.
- amelius 1y ago
  Maybe this is different for 10x versus 1x engineers?
mronetwo 1y ago
I get being optimistic but there’s a lot of ethical considerations that we’re choosing to ignore. The result is techno feudalism. Sure AI can help me with small things but it’s weird to be the guy preaching the gospel. In the end this is a product, sold by people who have more power than a single person ever had. They can do the marketing and hype, my interest lies in staying skeptical, especially in the incoming storm of AI generated misinformation or wave of students getting through university by cheating with AI.
- frankc 1y ago
  I dont think that is really true. There are many open weights models you can run yourself, including state of the art models like deepseek. Right now you its still expensive to run them at a reasonable speed, but for instance a $9500 mac studio can run deepseek at a reasonable, if not spectacular, speed.
  - nosianu 1y ago
    I would like to point to my comment in another AI thread not even a day ago: https://news.ycombinator.com/edit?id=44168022 https://news.ycombinator.com/edit?id=44168022 You can, just like you can theoretically use something else than Windows or MS Office. Until it's an entire ecosystem after a decade or two, and a workforce trained for that ecosystem, instead of right now easily substitutable AI providers. The current huge investments into the AI providers are not made because the investors are looking forward to everlasting fierce competition far beyond the initial stage.
    - the_mitsuhiko 1y ago
      > You can, just like you can theoretically use something else than Windows or MS Office. Until it's an entire ecosystem after a decade or two, and a workforce trained for that ecosystem, instead of right now easily substitutable AI providers. Author here: I had that thought for a while but I don't currently think that AI will unfold the same way (and I mentioned that in the post linked). I believe that at the speed this is going, and the innovation happening everywhere this will be a market with many models and players.
brador 1y ago
Using AI, the mental strain and toll of programming is reduced to what feels like 10% of usual. And that 10% is high level architecting and project planning. The most valuable 10x $ per hour time for a software engineer right now is in the shower, on the toilet, or just before shut eye. Not at the keyboard. Thinking time. Unfortunately for now and many phone time. The value of retaking that time is huge and growing.
ReptileMan 1y ago
One of the things I learned after starting using AI - (can anyone give information WTF are coding agents Google is flooded by info presented by cryptobros btw) is that I have really hated the whole typist part of the development. To the point where I prefer to make 2 prompts to the AI instead of modifying something myself even though it will take comparable time. Anyway the way it is going I was convinced that the oldest profession will be the only left standing when the AI runs its course in couple of decades. Now I am not sure even in its survival. Edit: We will know we are in the endgame when AI can reverse engineer a piece of software just by observing it's inputs and outputs. Like a game. Or crack Denuvo
- zihotki 1y ago
  > Or crack Denuvo That would be a nice benchmark!
xrisk 1y ago
> conviction that this future will be bright and worth embracing This is a bold claim and one that needs backing up.
intended 1y ago
You, the producer of an intellectual good, are searching for a buyer for your output and your labor. You, the consumer of intellectual good, have a machine that gives you average output. Most of the conversations of AI which worry about the impact, become easier when the producer and consumer positions are split and separated. If the tool is as good as the Hype claims (a high bar), then there is smaller space for human and human price matching. This gets further complicated, when we bring in the impact over time.
js8 1y ago
I see LLMs (AI if you will) as both regression and progression (and that's why I think people are so divided on it). It's progression for obvious reasons - it automates natural language processing (human language), which even 5 years ago was an unsolved problem. However, the regression part is less obvious. I think human progress, since enlightenment, was marked by increasingly externalizing human thoughts and judgement. That is, instead of relying on judgement of an authority (or intuition, lived and anecdotal experience, etc.), we increasingly demanded transparency and rigour in our thinking. This was facilitated by printing press and development of scientific method, but is seen everywhere (to the extent even philosophy split into analytical branch). However, with LLMs, we return to internalized thinking, based on judgement, this time of an algorithm. There is no way extract formal argument from an LLM, and so no way to ensure correctness in some sense. Whether you're an adopter or a skeptic depends where you stand on this issue (in a way, it's similar to type checking debate in programming). Personally, I think the correct way to use LLMs is to formalize the problem and then solve that using formal methods.
jordwest 1y ago
Great article, I also have no doubt we're in for some serious changes in society. But I disagree with the conclusion. I think the ideal place to be is neither optimistic nor pessimistic, but neutral. A conviction that the future is bright isn't enough to actually make the future bright. I was optimistic when Facebook first came out, and it turned into a dystopian nightmare. I was optimistic when self checkouts appeared, and now I feel like I'm a criminal by default every time I get groceries. I was optimistic about more people being online, and now we all feel like we can't get off it. I've become increasingly cynical about tech because of these and a thousand other disappointments about the utopia that was promised turning into not much more than an efficient means of extracting money from all of us. I don't think it's ever a good idea to be totally optimistic or pessimistic about a new technology, because then we're liable to ignore anything that goes against our narrative. Really, we don't know, and we can't know what's best for ourselves, usually until it's too late.
- _petronius 1y ago
  I have been thinking a lot about this parable (introduced to me by Bluey) in relation to the current surge of GenAI https://btribble.com/well-see/ https://btribble.com/well-see/ > A conviction that the future is bright isn't enough to actually make the future bright. Well said. We have to be curious, open, and ready to use the benefits of technological change, but also skeptical, concerned, and willing to defend social and political norms, too. Wild optimism is potentially very dangerous; hardline pessimism would leave us insufficiently ready to face the moment when we do need to act proactively.
  - jordwest 1y ago
    I love that parable. It stands in such contrast to a world where we all feel a pressure to know what’s right and wrong, how things will go etc. The reality is none of us do.
    - Karrot_Kream 1y ago
      I'm curious where you think this pressure comes from. I don't necessarily feel it myself but a lot of message board content I consume seems to react to this pressure, a pressure to have conviction in right and wrong. Is there something in the zeitgeist you feel that pushes people to have a strong stance?
000ooo000 1y ago
I can't help but feel these hyper optimistic takes on AI are just pure naïveté. Does one really think that redirecting the earning potential of a good chunk of the white collar population into the pockets of the already-wealthy is going to end well? I can't see a soft-landing scenario.
- matltc 1y ago
  There is a good chance this creates greater wealth disparity, and it seems certain that the white-collar job market will contract in the near-term. I am not sure that earning potential will necessarily be limited by this, however; the bar to do things has drastically been lowered for anyone with a computer and $200 a year to spend on a subscription to Claude or the like. Earning potential may increase, even for the displaced, in the long run. Still early
  - squishington 1y ago
    Earning potential should be considered in the context of increased wealth disparity though. After all, the rich and the poor are competing for the same resources. I work for a well known chip design company as an embedded software developer, and wages in my country (Australia) have not kept up with house prices (like many countries). Part of the housing price equation is buying power of the wealthy, which prices out younger people and those without family wealth. There's no scenario where my wages get me out of the rent cycle, no matter how productive I am and how much unpaid overtime I do to further the company goals.
    - matltc 1y ago
      I agree that the purchasing power of wages ought to be considered in this context. It is certainly true that one's wages must increase at a faster rate than prices for it to be considered a net positive. I would argue that housing prices are artificially inflated. Supply is constrained by regulations like zoning and rent controls, which causes scarcity and hence higher prices for existing housing. I would argue that, like other technological advances, the advent of AI will lead to greater productivity, and thus more abundant resources, given that the supply of labor does not shrink. Agree that housing is definitely an issue, but not one without a solution.
bachmeier 1y ago
> Do I program any faster? Not really. But it feels like I've gained 30% more time in my day because the machine is doing the work. I alternate between giving it instructions, reading a book, and reviewing the changes. If you would have told me even just six months ago that I'd prefer being an engineering lead to a virtual programmer intern over hitting the keys myself, I would not have believed it. I can go can make a coffee, and progress still happens. I can be at the playground with my youngest while work continues in the background. By this description, the economic case for AI is quite limited. I use AI all the time and for some things it is incredible. It has real downsides (getting things wrong, wasting time due to communication, size of prompt, context limits, etc.) that mean everything needs a thorough review. It also has financial costs. Benefits? Sure. A big shift for "everything"? Not remotely close to email or internet search. Let's keep some perspective. It's massive speculation to call AI a revolution based on current LLMs.
- the_mitsuhiko 1y ago
  > By this description, the economic case for AI is quite limited. If you do not assume concurrency that is absolutely correct. But even the Claude docs are pointing out that you can run more than one agent in parallel and that is really where the economic benefits would start to kick in.
  - d4rkn0d3z 1y ago
    How much does it cost to run several agents?
    - bdangubic 1y ago
      that is the wrong question, the right question is whether the cost is justified by benefits and the answer to (in the hands of right people) is yes
      - d4rkn0d3z 1y ago
        You cannot ask this follow-up unless you first answer mine as to cost. And not cost today but cost for the long haul.
_pdp_ 1y ago
> Do I program any faster? Not really. But it feels like I've gained 30% more time in my day because the machine is doing the work. This makes no sense. Besides, this is sort of written confession / admission that AI in this instance is as good as the OP's skills and the obvious outcome will follow soon. Personally I don't see AI as means to cut work by 30% but to do more with less.
- afro88 1y ago
  He means 30% of the time is waiting around for agents to reach a conclusion and need human review and a next task to tackle.
r9295 1y ago
How is one supposed to context switch between reading a book and rigorously understanding and verifying potentially hundreds of lines of code generated by an AI?
- afro88 1y ago
  The same way you context switch between whatever your day to day is and reviewing peers code
Quenby 1y ago
[dead]
saubeidl 1y ago
AI changes everything... for the worse. Of course it'll change things. But not a single thing will be better as a result, unless you're Mark Zuckerberg or one of his peers.
d4rkn0d3z 1y ago
I think if you stop labelling it "AI" and instead use "meta-software" or "natural language programming tools", then all of the hidden and externalized costs resurface. And a better understanding of skepticism arises. These tools may cost $20/month now but once all of the tooling to write software manually vanishes or becomes rarified, will the price stay the same? And will not the purveyors of this meta-software have a new form of control? Will intelligence itself be somehow sequestered? Probably not because what "AI" is giving you is not intelligence. Intelligence involves realizing that all human knowledge is fiction, careful, logical, predictive fiction. And "intelligence" involves remembering in exquisite detail what makes your knowledge fiction, so that you do not inadvertently become convinced that your careful logical predictive fiction is "truth". Predictions about "AI" involve various super exponentials combined and if you leave out just one of those that pulls in the wrong direction then the loose renormalization performed by wishful thinking fails. So rather than an "AI" programming utopia we end up in the same old efforts to manifest hegemony over computing, will to power. One algorythm to rule them all...the more things change the more they stay the same.
- Karrot_Kream 1y ago
  You can, today, use open source agents on open-weight models. You can point aider to locally running Gemma3 and no purveyor can stop you.
DarkCrusader2 1y ago
> Where I used to spend most of my time in Cursor, I now mostly use Claude Code, almost entirely hands-off. Do I program any faster? Not really. But it feels like I've gained 30% more time in my day because the machine is doing the work. I alternate between giving it instructions, reading a book, and reviewing the changes. This is very interesting to me and I did not know that our current LLMs can do this. Does anyone has a live coding video/stream of this workflow to do some non trivial task successfully, like developing/merging a feature for some open-source project? There is so much crap around AI on youtube that I have just stopped engaging with that stuff and not sure who are producing genuine content and who is just selling their latest grift.
cubefox 1y ago
In my frank opinion this optimism is simply, sorry, stupid. These tools will not stop at where they are. They will get better. They will more and more outperform more and more humans. They will become both more general (as opposed to good at some narrow tasks like writing text) and more capable. They will become "smarter than humans", first a little smarter in some respects, then much smarter on more respects. Eventually they will become superintelligent, in the same way in which we are superintelligent relative to a chimpanzee or an an ant. We will lose our jobs. Human labor will increasingly become worthless. We will be more hindrance than help, similar to how a monkey can't earn small amounts of money. His limited intelligence is not of limited use, instead it has negative value in any real work environment. But things will not stop at mass unemployment. People (e.g.) in Europe will not receive a significant UBI when the taxable AI companies are all US or Chinese companies. As wages tend to zero, people without investments in stocks or land will be increasingly impoverished compared to the rest. There will be an arms race, likely between the US an China. Other countries without big AI companies will be left in the dust. Each country big in AI will be incentivized to push forward as fast and recklessly as possible, security be damned. The Manhattan or Apollo programs will be a joke in comparison. Because only the country that achieves true superintelligence first will get a chance of controlling the future. But the question is not so much whether the US or China wins this arms race. The question is whether the resulting superintelligent system(s) will be controllable at all in the long term. They might escape human control relatively quickly once they significantly outperform us in all cognitive abilities. Similar to how animals can't control humans and children aren't in charge of their parents. And even if we somehow solve the problem of controlling things that are vastly smarter than us: In the long term, gradual disempowerment awaits. We will voluntarily offload more and more tasks to AI, more and more decisions, because anything else will become increasingly inefficient. And at some point we will realize that we have lost control, silently and forever, quite a while ago. The point of no return would be invisible. And eventually, the coming AI race may simply get rid of us, not because of maliciousness, but because we are in the way. In the way of projects too large for us to comprehend, similar to how no animal can begin to comprehend why its habitat is getting destroyed by human urbanization. The only slim hope is that superintelligence will be created in such a way which ensures that they want to care for us, like some people genuinely care for their favorite pets. Then we will get our AI utopia, then our future will be bright. But I wouldn't bet on it. Yet all these considerations are readily ignored. Most people can't or refuse to extrapolate the exponential AI growth of the last years. Or they think AI will "hit a wall" slightly before superhuman intelligence. Which is astonishingly stupid. No way say it in any other way. It's all so obvious. It's all so obvious.
vdupras 1y ago
Articles like this makes this uneasy though resurface: something is wrong with this path. It's not really about AI itself, but about the efficiency of the path. AI proponents tell us that this tool enable them to be purely creative by expressing their ideas without having any of the churn associated with the implementation. But rather than telling us that we "need" AI, doesn't it tell us that modern tools are mis-designed? Wouldn't it be possible to imagine a set of tools that is better designed and allow us to express creative ideas more purely, with less churn and, most importantly, without the uneasy "fuzziness" associated with AI?
pjio 1y ago
After decades of trying to make the code encapsulate as much meaning in as few words as possible now there seems to be a new trend of "more code is better" and because AI writes and reads it, it's ok. I'm already sick of finding subtle bugs in AI generated code, looking at fever dream like AI images and reading text without the intentional style which indicates a human writer.
incomingpain 1y ago
I will happily lose more HN score to agree. AI is amazing. Pycharm's AI i guess defaults to claude sonnet 3.5 even though 4 is available? Is that claude code? It has multiple options, i guess im happy with 3.5. It has ollama and lm studio options as well for those who need privacy. I got LM studio working yesterday but so far very bad; it desparately needs the ability to check the internet. >Do I program any faster? Not really. But it feels like I've gained 30% more time in my day because the machine is doing the work. I wonder if anyone has objectively tried to figure out if you code faster or not. >While all this is happening, I've found myself reflecting a lot on what AI means to the world and I am becoming increasingly optimistic about our future. I am so optimistic. The worker shortage of 2030 depends on no significant boost to productivity. But AI is going to be a significant boost to productivity. AI is likely going to enable us to automate multiple industries. Could we eliminate food scarcity by fully automating food production? >That's what makes this moment feel so strange — half revolutionary, half prelude. And yet, oddly, there are so many technologists who are holdouts. How could techies reject this change? I think it's a silent majority situation. The techies are all working on new projects at lightning speed with AI; too busy to post on social media. Whereas those without AI are complaining about AI.
roxolotl 1y ago
> We've built something beyond what we imagined, and instead of being curious, many are dismissive and denying its capabilities. What is that? I’ll keep saying this but the problem is the hype. The AI is Normal Technology[0] piece is the first piece that made me even remotely excited. The hype destroys everything positive about the technology because it totally fills the discussion space with nonsense. For many it feels like fruit from the poison tree both because the data was acquired using dubious means and the people selling AI are only interested in world domination. Of course it’s going to be transformative. Of course it’s a powerful tool. Of course it has limitations. Of course some people are going to make boat loads of money. Of course some people are going to lose a lot of money. Same as it was with many other technologies that changed everything. 0: https://knightcolumbia.org/content/ai-as-normal-technology https://knightcolumbia.org/content/ai-as-normal-technology
mdp2021 1y ago
> AI ... can accelerate innovation in medicine, science, and engineering. // Right now it's messy and raw With all due respect to Armin, he sounds confused. That sentence above amounts to a "[Software] can accelerate innovation in medicine, science, and engineering. // Right now it's messy and raw" - as I have to note again, after a piece hours ago from the MIT¹. The linear reply is "Dude, what are you talking about? ?! !!". ¹ https://news.ycombinator.com/item?id=44167355 https://news.ycombinator.com/item?id=44167355
ChrisArchitect 1y ago
> For sure the job of programmers and artists will change, but they won't vanish. But they will vanish for the vast majority. It's the scale of it. The scale of the vanishing is not miniscule. > People will always value well crafted products. But again, in this case what will be left is only a few crafters, the privileged few, because no one else can afford to delve into whatever space. It's the scale of the destruction. Lose me in these articles when they take on this sort of privileged engineer dismissiveness.

55 points by tosh 1y ago

I genuinely feel like we’re living through another big shift — like what smartphones did in the 2000s, AI is doing now. I’m not a developer, but I can already feel how the way I use the internet is changing. I used to search for things, now I just ask — and the AI replies like a person. It’s subtle, but powerful. I really think in a few years, most of us will have some kind of personal AI. Not just a tool, but more like a life companion — helping us plan, decide, learn, even reflect. Kind of like how we can’t imagine life without Google now.

inferiorhuman 1y ago
Yeah it's definitely changed the internet. All of sudden every site is behind anti-bot garbage (usually Cloudflare). Sites I use (like Sourcehut) are getting DDOS'd by AI scrapers. It might be a wee bit reductive to say that AI is destroying the internet, but this certainly is the death of the open internet. AI has just handed a huge amount of power to companies like Cloudflare and made it that much more difficult to avoid being tracked.

csomar 1y ago

> Do I program any faster? Not really. This really got my attention. I use Claude heavily and although I am sure it is saving me considerable time, I feel like my "delivery" speed is roughly the same. I think the realization that will come from this is that writing code is not that consuming after all; unless you are doing cryptography or math. Most of your time is spent testing, deploying, figuring out some weird bug, writing docs somewhere, signing up to some service, etc...

Frieren 1y ago
While writing code I think about what I am doing, improvements, etc. To write easy boilerplate code helps me to find improvements or even realize that there is some risk I overlooked. It is a zen state that helps with the deep thinking part. Not writing code forces me to read and reread the code to get the same realization. That is why it is harder to work with code written by other people. In the age of AI all code is code written by someone else that I have to maintain, it looks like a nightmare.
- fhd2 1y ago
  Yeah I have the same. In general, writing (and experimenting) is how I _think_. If I read code, it takes me a lot longer to understand. Sure, I save some time by not having to do any trial and error, I'm looking at a solution that already works according to some testing. But then I start to wonder about edge cases, leaky abstractions and such, and not having done the work, that's where a lot of the effort saved by not writing it comes back, at a stage where the work is seemingly already done and "just" needs reviewing, which is somehow more frustrating. Perhaps people are just different. I work great on a blank canvas. I know a lot of people struggle immensely with it. Hell, some people type so slow or have such low mastery of their tools, I really feel their pain watching them.
- amelius 1y ago
  Yes, AI is taking all the fun jobs.
- AlexeyBrin 1y ago
  One old trick that helps a bit with understanding, is to not copy paste what the AI suggested, but rather to manually copy the code suggested the way you would do it with a physical book.
the_mitsuhiko 1y ago
> Most of your time is spent testing, deploying I became hyper attentive to iteration speeds as a result. This is not just something a human runs into, the agent runs into it even more because it loves to run tests etc. The quicker that goes, the more efficient the agent is. Obviously the same should be true for humans but because the agent doesn't make breaks quite like a human does, it becomes more noticeable.
- csomar 1y ago
  It gets me frustrated. Here you are one second flying, in the zone, moving from concept to concept at faster than human speeds; and then boof, you have to pull up some random api doc, open postman, test the api, come back to the chat console, etc...
  - the_mitsuhiko 1y ago
    What's nice is that a codebase optimized for agent also becomes a codebase optimized for humans.
    - saubeidl 1y ago
      Disagree. A codebase optimized for agents is overly verbose and takes more conscious effort to parse.
      - the_mitsuhiko 1y ago
        Overly verbose codebases are not great for agents in my experience because they confuse them greatly and make refactoring hard. However codebases that I find to work really well with agents are quite lite on abstractions and I always felt that this is the right way to build software for humans too. I have a lot of functions, I pass a lot of data, I embrace a lot of thread local state and this works really well for both humans and AI to understand what is going on. I gave a talk on this recently (not the agentic view), but I found this to work exceptionally well for agents: https://www.youtube.com/watch?v=ej5RsTtVvQE https://www.youtube.com/watch?v=ej5RsTtVvQE But here is where agents and humans both agree: quick and easy tooling. Make it easy to iterate on a test, make it easy to manage resources, make the iteration cycle as quick as possible. I'm writing this as my partner's job forces iteration cycles of minutes on her. No agent can do better in that space than a human.
herbst 1y ago
I am undecided about AI coding. Ive done some vibe coding with working but code wise horribly results. But I have had working prototypes faster than ever. But I recently starting doing some redesign and architecture changes (manually) in an old code base and notice that i tab trough whole files auto fixing code one by one with suggestions which is amazing. I think we need to find a way to work with it and keep it far away from making architectural decisions for now.
meindnoch 1y ago
>writing code is not that consuming after all; unless you are doing cryptography or math Most cryptographic algorithms fit on a napkin. 99% of the work goes into proving the mathematical properties of the algorithm.
- yencabulator 1y ago
  But that napkin may not be constant time.
amelius 1y ago
Maybe this is different for 10x versus 1x engineers?

mronetwo 1y ago

I get being optimistic but there’s a lot of ethical considerations that we’re choosing to ignore. The result is techno feudalism. Sure AI can help me with small things but it’s weird to be the guy preaching the gospel. In the end this is a product, sold by people who have more power than a single person ever had. They can do the marketing and hype, my interest lies in staying skeptical, especially in the incoming storm of AI generated misinformation or wave of students getting through university by cheating with AI.

frankc 1y ago
I dont think that is really true. There are many open weights models you can run yourself, including state of the art models like deepseek. Right now you its still expensive to run them at a reasonable speed, but for instance a $9500 mac studio can run deepseek at a reasonable, if not spectacular, speed.
- nosianu 1y ago
  I would like to point to my comment in another AI thread not even a day ago: https://news.ycombinator.com/edit?id=44168022 https://news.ycombinator.com/edit?id=44168022 You can, just like you can theoretically use something else than Windows or MS Office. Until it's an entire ecosystem after a decade or two, and a workforce trained for that ecosystem, instead of right now easily substitutable AI providers. The current huge investments into the AI providers are not made because the investors are looking forward to everlasting fierce competition far beyond the initial stage.
  - the_mitsuhiko 1y ago
    > You can, just like you can theoretically use something else than Windows or MS Office. Until it's an entire ecosystem after a decade or two, and a workforce trained for that ecosystem, instead of right now easily substitutable AI providers. Author here: I had that thought for a while but I don't currently think that AI will unfold the same way (and I mentioned that in the post linked). I believe that at the speed this is going, and the innovation happening everywhere this will be a market with many models and players.

brador 1y ago

Using AI, the mental strain and toll of programming is reduced to what feels like 10% of usual. And that 10% is high level architecting and project planning. The most valuable 10x $ per hour time for a software engineer right now is in the shower, on the toilet, or just before shut eye. Not at the keyboard. Thinking time. Unfortunately for now and many phone time. The value of retaking that time is huge and growing.

ReptileMan 1y ago

One of the things I learned after starting using AI - (can anyone give information WTF are coding agents Google is flooded by info presented by cryptobros btw) is that I have really hated the whole typist part of the development. To the point where I prefer to make 2 prompts to the AI instead of modifying something myself even though it will take comparable time. Anyway the way it is going I was convinced that the oldest profession will be the only left standing when the AI runs its course in couple of decades. Now I am not sure even in its survival. Edit: We will know we are in the endgame when AI can reverse engineer a piece of software just by observing it's inputs and outputs. Like a game. Or crack Denuvo

zihotki 1y ago
> Or crack Denuvo That would be a nice benchmark!

xrisk 1y ago

> conviction that this future will be bright and worth embracing This is a bold claim and one that needs backing up.

intended 1y ago

You, the producer of an intellectual good, are searching for a buyer for your output and your labor. You, the consumer of intellectual good, have a machine that gives you average output. Most of the conversations of AI which worry about the impact, become easier when the producer and consumer positions are split and separated. If the tool is as good as the Hype claims (a high bar), then there is smaller space for human and human price matching. This gets further complicated, when we bring in the impact over time.

js8 1y ago

I see LLMs (AI if you will) as both regression and progression (and that's why I think people are so divided on it). It's progression for obvious reasons - it automates natural language processing (human language), which even 5 years ago was an unsolved problem. However, the regression part is less obvious. I think human progress, since enlightenment, was marked by increasingly externalizing human thoughts and judgement. That is, instead of relying on judgement of an authority (or intuition, lived and anecdotal experience, etc.), we increasingly demanded transparency and rigour in our thinking. This was facilitated by printing press and development of scientific method, but is seen everywhere (to the extent even philosophy split into analytical branch). However, with LLMs, we return to internalized thinking, based on judgement, this time of an algorithm. There is no way extract formal argument from an LLM, and so no way to ensure correctness in some sense. Whether you're an adopter or a skeptic depends where you stand on this issue (in a way, it's similar to type checking debate in programming). Personally, I think the correct way to use LLMs is to formalize the problem and then solve that using formal methods.

jordwest 1y ago

Great article, I also have no doubt we're in for some serious changes in society. But I disagree with the conclusion. I think the ideal place to be is neither optimistic nor pessimistic, but neutral. A conviction that the future is bright isn't enough to actually make the future bright. I was optimistic when Facebook first came out, and it turned into a dystopian nightmare. I was optimistic when self checkouts appeared, and now I feel like I'm a criminal by default every time I get groceries. I was optimistic about more people being online, and now we all feel like we can't get off it. I've become increasingly cynical about tech because of these and a thousand other disappointments about the utopia that was promised turning into not much more than an efficient means of extracting money from all of us. I don't think it's ever a good idea to be totally optimistic or pessimistic about a new technology, because then we're liable to ignore anything that goes against our narrative. Really, we don't know, and we can't know what's best for ourselves, usually until it's too late.

_petronius 1y ago
I have been thinking a lot about this parable (introduced to me by Bluey) in relation to the current surge of GenAI https://btribble.com/well-see/ https://btribble.com/well-see/ > A conviction that the future is bright isn't enough to actually make the future bright. Well said. We have to be curious, open, and ready to use the benefits of technological change, but also skeptical, concerned, and willing to defend social and political norms, too. Wild optimism is potentially very dangerous; hardline pessimism would leave us insufficiently ready to face the moment when we do need to act proactively.
- jordwest 1y ago
  I love that parable. It stands in such contrast to a world where we all feel a pressure to know what’s right and wrong, how things will go etc. The reality is none of us do.
  - Karrot_Kream 1y ago
    I'm curious where you think this pressure comes from. I don't necessarily feel it myself but a lot of message board content I consume seems to react to this pressure, a pressure to have conviction in right and wrong. Is there something in the zeitgeist you feel that pushes people to have a strong stance?

000ooo000 1y ago

I can't help but feel these hyper optimistic takes on AI are just pure naïveté. Does one really think that redirecting the earning potential of a good chunk of the white collar population into the pockets of the already-wealthy is going to end well? I can't see a soft-landing scenario.

matltc 1y ago
There is a good chance this creates greater wealth disparity, and it seems certain that the white-collar job market will contract in the near-term. I am not sure that earning potential will necessarily be limited by this, however; the bar to do things has drastically been lowered for anyone with a computer and $200 a year to spend on a subscription to Claude or the like. Earning potential may increase, even for the displaced, in the long run. Still early
- squishington 1y ago
  Earning potential should be considered in the context of increased wealth disparity though. After all, the rich and the poor are competing for the same resources. I work for a well known chip design company as an embedded software developer, and wages in my country (Australia) have not kept up with house prices (like many countries). Part of the housing price equation is buying power of the wealthy, which prices out younger people and those without family wealth. There's no scenario where my wages get me out of the rent cycle, no matter how productive I am and how much unpaid overtime I do to further the company goals.
  - matltc 1y ago
    I agree that the purchasing power of wages ought to be considered in this context. It is certainly true that one's wages must increase at a faster rate than prices for it to be considered a net positive. I would argue that housing prices are artificially inflated. Supply is constrained by regulations like zoning and rent controls, which causes scarcity and hence higher prices for existing housing. I would argue that, like other technological advances, the advent of AI will lead to greater productivity, and thus more abundant resources, given that the supply of labor does not shrink. Agree that housing is definitely an issue, but not one without a solution.

bachmeier 1y ago

> Do I program any faster? Not really. But it feels like I've gained 30% more time in my day because the machine is doing the work. I alternate between giving it instructions, reading a book, and reviewing the changes. If you would have told me even just six months ago that I'd prefer being an engineering lead to a virtual programmer intern over hitting the keys myself, I would not have believed it. I can go can make a coffee, and progress still happens. I can be at the playground with my youngest while work continues in the background. By this description, the economic case for AI is quite limited. I use AI all the time and for some things it is incredible. It has real downsides (getting things wrong, wasting time due to communication, size of prompt, context limits, etc.) that mean everything needs a thorough review. It also has financial costs. Benefits? Sure. A big shift for "everything"? Not remotely close to email or internet search. Let's keep some perspective. It's massive speculation to call AI a revolution based on current LLMs.

the_mitsuhiko 1y ago
> By this description, the economic case for AI is quite limited. If you do not assume concurrency that is absolutely correct. But even the Claude docs are pointing out that you can run more than one agent in parallel and that is really where the economic benefits would start to kick in.
- d4rkn0d3z 1y ago
  How much does it cost to run several agents?
  - bdangubic 1y ago
    that is the wrong question, the right question is whether the cost is justified by benefits and the answer to (in the hands of right people) is yes
    - d4rkn0d3z 1y ago
      You cannot ask this follow-up unless you first answer mine as to cost. And not cost today but cost for the long haul.

_pdp_ 1y ago

> Do I program any faster? Not really. But it feels like I've gained 30% more time in my day because the machine is doing the work. This makes no sense. Besides, this is sort of written confession / admission that AI in this instance is as good as the OP's skills and the obvious outcome will follow soon. Personally I don't see AI as means to cut work by 30% but to do more with less.

afro88 1y ago
He means 30% of the time is waiting around for agents to reach a conclusion and need human review and a next task to tackle.

r9295 1y ago

How is one supposed to context switch between reading a book and rigorously understanding and verifying potentially hundreds of lines of code generated by an AI?

afro88 1y ago
The same way you context switch between whatever your day to day is and reviewing peers code

Quenby 1y ago

[dead]

saubeidl 1y ago

AI changes everything... for the worse. Of course it'll change things. But not a single thing will be better as a result, unless you're Mark Zuckerberg or one of his peers.

d4rkn0d3z 1y ago

I think if you stop labelling it "AI" and instead use "meta-software" or "natural language programming tools", then all of the hidden and externalized costs resurface. And a better understanding of skepticism arises. These tools may cost $20/month now but once all of the tooling to write software manually vanishes or becomes rarified, will the price stay the same? And will not the purveyors of this meta-software have a new form of control? Will intelligence itself be somehow sequestered? Probably not because what "AI" is giving you is not intelligence. Intelligence involves realizing that all human knowledge is fiction, careful, logical, predictive fiction. And "intelligence" involves remembering in exquisite detail what makes your knowledge fiction, so that you do not inadvertently become convinced that your careful logical predictive fiction is "truth". Predictions about "AI" involve various super exponentials combined and if you leave out just one of those that pulls in the wrong direction then the loose renormalization performed by wishful thinking fails. So rather than an "AI" programming utopia we end up in the same old efforts to manifest hegemony over computing, will to power. One algorythm to rule them all...the more things change the more they stay the same.

Karrot_Kream 1y ago
You can, today, use open source agents on open-weight models. You can point aider to locally running Gemma3 and no purveyor can stop you.

DarkCrusader2 1y ago

> Where I used to spend most of my time in Cursor, I now mostly use Claude Code, almost entirely hands-off. Do I program any faster? Not really. But it feels like I've gained 30% more time in my day because the machine is doing the work. I alternate between giving it instructions, reading a book, and reviewing the changes. This is very interesting to me and I did not know that our current LLMs can do this. Does anyone has a live coding video/stream of this workflow to do some non trivial task successfully, like developing/merging a feature for some open-source project? There is so much crap around AI on youtube that I have just stopped engaging with that stuff and not sure who are producing genuine content and who is just selling their latest grift.

cubefox 1y ago

In my frank opinion this optimism is simply, sorry, stupid. These tools will not stop at where they are. They will get better. They will more and more outperform more and more humans. They will become both more general (as opposed to good at some narrow tasks like writing text) and more capable. They will become "smarter than humans", first a little smarter in some respects, then much smarter on more respects. Eventually they will become superintelligent, in the same way in which we are superintelligent relative to a chimpanzee or an an ant. We will lose our jobs. Human labor will increasingly become worthless. We will be more hindrance than help, similar to how a monkey can't earn small amounts of money. His limited intelligence is not of limited use, instead it has negative value in any real work environment. But things will not stop at mass unemployment. People (e.g.) in Europe will not receive a significant UBI when the taxable AI companies are all US or Chinese companies. As wages tend to zero, people without investments in stocks or land will be increasingly impoverished compared to the rest. There will be an arms race, likely between the US an China. Other countries without big AI companies will be left in the dust. Each country big in AI will be incentivized to push forward as fast and recklessly as possible, security be damned. The Manhattan or Apollo programs will be a joke in comparison. Because only the country that achieves true superintelligence first will get a chance of controlling the future. But the question is not so much whether the US or China wins this arms race. The question is whether the resulting superintelligent system(s) will be controllable at all in the long term. They might escape human control relatively quickly once they significantly outperform us in all cognitive abilities. Similar to how animals can't control humans and children aren't in charge of their parents. And even if we somehow solve the problem of controlling things that are vastly smarter than us: In the long term, gradual disempowerment awaits. We will voluntarily offload more and more tasks to AI, more and more decisions, because anything else will become increasingly inefficient. And at some point we will realize that we have lost control, silently and forever, quite a while ago. The point of no return would be invisible. And eventually, the coming AI race may simply get rid of us, not because of maliciousness, but because we are in the way. In the way of projects too large for us to comprehend, similar to how no animal can begin to comprehend why its habitat is getting destroyed by human urbanization. The only slim hope is that superintelligence will be created in such a way which ensures that they want to care for us, like some people genuinely care for their favorite pets. Then we will get our AI utopia, then our future will be bright. But I wouldn't bet on it. Yet all these considerations are readily ignored. Most people can't or refuse to extrapolate the exponential AI growth of the last years. Or they think AI will "hit a wall" slightly before superhuman intelligence. Which is astonishingly stupid. No way say it in any other way. It's all so obvious. It's all so obvious.

vdupras 1y ago

Articles like this makes this uneasy though resurface: something is wrong with this path. It's not really about AI itself, but about the efficiency of the path. AI proponents tell us that this tool enable them to be purely creative by expressing their ideas without having any of the churn associated with the implementation. But rather than telling us that we "need" AI, doesn't it tell us that modern tools are mis-designed? Wouldn't it be possible to imagine a set of tools that is better designed and allow us to express creative ideas more purely, with less churn and, most importantly, without the uneasy "fuzziness" associated with AI?

pjio 1y ago

After decades of trying to make the code encapsulate as much meaning in as few words as possible now there seems to be a new trend of "more code is better" and because AI writes and reads it, it's ok. I'm already sick of finding subtle bugs in AI generated code, looking at fever dream like AI images and reading text without the intentional style which indicates a human writer.

incomingpain 1y ago

I will happily lose more HN score to agree. AI is amazing. Pycharm's AI i guess defaults to claude sonnet 3.5 even though 4 is available? Is that claude code? It has multiple options, i guess im happy with 3.5. It has ollama and lm studio options as well for those who need privacy. I got LM studio working yesterday but so far very bad; it desparately needs the ability to check the internet. >Do I program any faster? Not really. But it feels like I've gained 30% more time in my day because the machine is doing the work. I wonder if anyone has objectively tried to figure out if you code faster or not. >While all this is happening, I've found myself reflecting a lot on what AI means to the world and I am becoming increasingly optimistic about our future. I am so optimistic. The worker shortage of 2030 depends on no significant boost to productivity. But AI is going to be a significant boost to productivity. AI is likely going to enable us to automate multiple industries. Could we eliminate food scarcity by fully automating food production? >That's what makes this moment feel so strange — half revolutionary, half prelude. And yet, oddly, there are so many technologists who are holdouts. How could techies reject this change? I think it's a silent majority situation. The techies are all working on new projects at lightning speed with AI; too busy to post on social media. Whereas those without AI are complaining about AI.

roxolotl 1y ago

> We've built something beyond what we imagined, and instead of being curious, many are dismissive and denying its capabilities. What is that? I’ll keep saying this but the problem is the hype. The AI is Normal Technology[0] piece is the first piece that made me even remotely excited. The hype destroys everything positive about the technology because it totally fills the discussion space with nonsense. For many it feels like fruit from the poison tree both because the data was acquired using dubious means and the people selling AI are only interested in world domination. Of course it’s going to be transformative. Of course it’s a powerful tool. Of course it has limitations. Of course some people are going to make boat loads of money. Of course some people are going to lose a lot of money. Same as it was with many other technologies that changed everything. 0: https://knightcolumbia.org/content/ai-as-normal-technology https://knightcolumbia.org/content/ai-as-normal-technology

mdp2021 1y ago

> AI ... can accelerate innovation in medicine, science, and engineering. // Right now it's messy and raw With all due respect to Armin, he sounds confused. That sentence above amounts to a "[Software] can accelerate innovation in medicine, science, and engineering. // Right now it's messy and raw" - as I have to note again, after a piece hours ago from the MIT¹. The linear reply is "Dude, what are you talking about? ?! !!". ¹ https://news.ycombinator.com/item?id=44167355 https://news.ycombinator.com/item?id=44167355

ChrisArchitect 1y ago

> For sure the job of programmers and artists will change, but they won't vanish. But they will vanish for the vast majority. It's the scale of it. The scale of the vanishing is not miniscule. > People will always value well crafted products. But again, in this case what will be left is only a few crafters, the privileged few, because no one else can afford to delve into whatever space. It's the scale of the destruction. Lose me in these articles when they take on this sort of privileged engineer dismissiveness.