7 ms·

I had my "wow" moment recently and saw an AI agent do something do over a weekend something that I spent years developing a relatively niche skill might have ne

I had my "wow" moment recently and saw an AI agent do something do over a weekend something that I spent years developing a relatively niche skill might have needed a month to complete. I think the message the team lead takes away right now is "oh, this can be done a lot faster, let's do more!" This is exciting for a bit, but I think it comes at the cost of learning -- and being able to communicate --architectural lessons that would stage off systemic issues later.

embedding-shape 3mo ago
My personal "wow" moment was less about it being able to do the same that I could do with it, just faster, but instead that together we could could build things that I wasn't able to build before (or unwilling to spend the time to learn maybe). So a bunch of stuff are suddenly "unlocked", as long as you know how to verify it properly.
- timbert 3mo ago
  ^ This! I have been able to build applications that I couldn't dream of before, not because of a lack of capability on my part, but because of time: I don't have the time to learn all of the new APIs and frameworks to accomplish my goal. Now I can build far more complex useful applications, with functions and features that would take me months or more to learn how to code. The downside is that I didn't learn how to use those APIs, but it can be argued that these skills are low value to begin with. APIs change, there are competing APIs that do the same thing, and you may only use that skill for one project and then move on. Gaining those skills can be a big time sink and may not result in long-term gain. Think of all the time spent learning the ins and outs of come-and-gone APIs and frameworks over the decades (I chose not to provide examples to save the resulting side-arguments, but I know you can rattle some off in your head). Those APIs and frameworks were bridges to make developing easier for the common developer, but they change so quickly that I'd argue they waste our time, and lead to almost endless re-work as we re-architect our software for the framework de jour. And they grow in complexity over time, ultimately becoming the next bloated framework. It's an endless, wasteful cycle. Maybe AI can help us get out of that.
  - _DeadFred_ 3mo ago
    Sounds like you appreciate the part where AI is an entropy machine.
- bluefirebrand 3mo ago
  If you weren't able to build it before then you aren't able to properly verify it Verifying is harder than building!
  - embedding-shape 3mo ago
    Ok, so before I couldn't (/didnt have the time to) build cross-platform native desktop applications with minimal amount of 3rd party dependencies, but now I "can". What exactly am I unable to verify here?
    - bluefirebrand 3mo ago
      Couldn't is "didn't have the skills to" That's pretty different from "didn't have the time/money/motivation to" If you didn't have the skills to build something before then you don't have the skills to verify something is correct after Clicking around a finished UI is not verifying
      - embedding-shape 3mo ago
        > That's pretty different from "didn't have the time/money/motivation to" Maybe start at the first message here: https://news.ycombinator.com/item?id=48553444 https://news.ycombinator.com/item?id=48553444 I still don't have the skill/knowledge to build cross-platform native applications the way I'm doing it with an agent now. I am still able to verify it's correct. How is that possible if what you say is correct? > Clicking around a finished UI is not verifying Definitively a part of it, if you care about having mouse interactions and having graphical UIs.
  - rpdillon 3mo ago
    Verifying is harder than building for some problems, but not all. I think part of the trick of using AI is applying it where verification is easy.
  - sameerds 3mo ago
    > If you weren't able to build it before then you aren't able to properly verify it GP said "couldn't dream of building". To make an analogy, I wouldn't dream of building a house if I had to do every bit of brick-laying myself. But with others doing what I can't do, I can indeed dream of building a house and I do know how to verify that the house was built to satisfy my dream.
- WarmWash 3mo ago
  There was a period where computer+grandmaster was the ultimate winning force in chess. It lasted a few years before it was just computer.
- thrixton 3mo ago
  Absolutely. Things which used to take me a week to grind through, finding edge cases, trial and error, researching online, now takes an afternoon directing an agent and verifying the solution. For a generalist, it's a huge force multiplier.
  - b112 3mo ago
    Wonderful now. In the future, there will be no more generalists with your capabilities. At all. No one will ever develop the skill. We're at the end of our career's lifetime. We're like the horse watching more and more cars appearing. Within a generation, the coding languages we use today will all vanish, to be replaced with direct to metal methods, assembly, etc. The only reason C exists, is for humans. PHP, rust, python is for humans. There will be no more compilers. No more human-readable languages. No more scripting with readable code. No more open or closed source. Imagine taking every aspect of computing which is designed for humans to understand, and removing it, because it's redundant. Now once removed, and once humans no longer understand on their own, imagine trying to build something independently? Imagine trying to bootstrap modern compute, without all the layers which came before? My point is that soon, perhaps 20 years, most humans will be incapable of coding, developing, or working with compute in any way outside of an end-user role. Even those previously skilled, will now be looking at an entire kernel written in assembly. And that's before new CPUs appear, with new assembly. New assembly, once again, not designed by or for humans to understand. And yes, current assembly is designed that way. We are not simply looking at a bit of help here. We're looking at the entire end of every single computing job, ever. We're looking at the complete loss of control, of something which is the underpinning of our entire society. Every aspect of everything we have, runs on a chip now. Personal transport, everything in health care, everything in resource extraction, everything harvesting food through to delivery to table, all of it involves some aspect of compute. And we soon will not comprehend nor create that underlying aspect of society. I dunno. It all seems like the answer to the fermi paradox.
spicyusername 3mo ago
Everyone is afraid about what will happen when people's skills atrophy, but the tools aren't going anywhere. If they truly can always recognize problems and communicate them in understandable ways, then people's skills won't atrophy. Everyone's skills will be enhanced. New developers will learn all of the things they need to learn through using the tool.
- izzydata 3mo ago
  The tools can definitely go away. They could get exponentially more expensive to where it is no longer financially possible to keep it subsidized. The bubble could pop and most of the companies could go bankrupt. It's not exactly like a calculator because it requires such an astronomical amount of resources to keep running.
  - djmips 3mo ago
    I agree with you but there is a calculator analogy where the resources are local and while taxing for most, viable for the individual. Local AI is less capable than an army frontier model agents but still quite useful.
- theodric 3mo ago
  I think the most obvious impact (good AND bad) will be from people who didn't have (or want) skills to begin with being empowered to have a model crank out custom software to do little things to make their lives better. Can't atrophy a muscle that doesn't exist!
kobe_bryant 3mo ago
feel free to share even a single detail

embedding-shape 3mo ago

My personal "wow" moment was less about it being able to do the same that I could do with it, just faster, but instead that together we could could build things that I wasn't able to build before (or unwilling to spend the time to learn maybe). So a bunch of stuff are suddenly "unlocked", as long as you know how to verify it properly.

timbert 3mo ago
^ This! I have been able to build applications that I couldn't dream of before, not because of a lack of capability on my part, but because of time: I don't have the time to learn all of the new APIs and frameworks to accomplish my goal. Now I can build far more complex useful applications, with functions and features that would take me months or more to learn how to code. The downside is that I didn't learn how to use those APIs, but it can be argued that these skills are low value to begin with. APIs change, there are competing APIs that do the same thing, and you may only use that skill for one project and then move on. Gaining those skills can be a big time sink and may not result in long-term gain. Think of all the time spent learning the ins and outs of come-and-gone APIs and frameworks over the decades (I chose not to provide examples to save the resulting side-arguments, but I know you can rattle some off in your head). Those APIs and frameworks were bridges to make developing easier for the common developer, but they change so quickly that I'd argue they waste our time, and lead to almost endless re-work as we re-architect our software for the framework de jour. And they grow in complexity over time, ultimately becoming the next bloated framework. It's an endless, wasteful cycle. Maybe AI can help us get out of that.
- _DeadFred_ 3mo ago
  Sounds like you appreciate the part where AI is an entropy machine.
bluefirebrand 3mo ago
If you weren't able to build it before then you aren't able to properly verify it Verifying is harder than building!
- embedding-shape 3mo ago
  Ok, so before I couldn't (/didnt have the time to) build cross-platform native desktop applications with minimal amount of 3rd party dependencies, but now I "can". What exactly am I unable to verify here?
  - bluefirebrand 3mo ago
    Couldn't is "didn't have the skills to" That's pretty different from "didn't have the time/money/motivation to" If you didn't have the skills to build something before then you don't have the skills to verify something is correct after Clicking around a finished UI is not verifying
    - embedding-shape 3mo ago
      > That's pretty different from "didn't have the time/money/motivation to" Maybe start at the first message here: https://news.ycombinator.com/item?id=48553444 https://news.ycombinator.com/item?id=48553444 I still don't have the skill/knowledge to build cross-platform native applications the way I'm doing it with an agent now. I am still able to verify it's correct. How is that possible if what you say is correct? > Clicking around a finished UI is not verifying Definitively a part of it, if you care about having mouse interactions and having graphical UIs.
- rpdillon 3mo ago
  Verifying is harder than building for some problems, but not all. I think part of the trick of using AI is applying it where verification is easy.
- sameerds 3mo ago
  > If you weren't able to build it before then you aren't able to properly verify it GP said "couldn't dream of building". To make an analogy, I wouldn't dream of building a house if I had to do every bit of brick-laying myself. But with others doing what I can't do, I can indeed dream of building a house and I do know how to verify that the house was built to satisfy my dream.
WarmWash 3mo ago
There was a period where computer+grandmaster was the ultimate winning force in chess. It lasted a few years before it was just computer.
thrixton 3mo ago
Absolutely. Things which used to take me a week to grind through, finding edge cases, trial and error, researching online, now takes an afternoon directing an agent and verifying the solution. For a generalist, it's a huge force multiplier.
- b112 3mo ago
  Wonderful now. In the future, there will be no more generalists with your capabilities. At all. No one will ever develop the skill. We're at the end of our career's lifetime. We're like the horse watching more and more cars appearing. Within a generation, the coding languages we use today will all vanish, to be replaced with direct to metal methods, assembly, etc. The only reason C exists, is for humans. PHP, rust, python is for humans. There will be no more compilers. No more human-readable languages. No more scripting with readable code. No more open or closed source. Imagine taking every aspect of computing which is designed for humans to understand, and removing it, because it's redundant. Now once removed, and once humans no longer understand on their own, imagine trying to build something independently? Imagine trying to bootstrap modern compute, without all the layers which came before? My point is that soon, perhaps 20 years, most humans will be incapable of coding, developing, or working with compute in any way outside of an end-user role. Even those previously skilled, will now be looking at an entire kernel written in assembly. And that's before new CPUs appear, with new assembly. New assembly, once again, not designed by or for humans to understand. And yes, current assembly is designed that way. We are not simply looking at a bit of help here. We're looking at the entire end of every single computing job, ever. We're looking at the complete loss of control, of something which is the underpinning of our entire society. Every aspect of everything we have, runs on a chip now. Personal transport, everything in health care, everything in resource extraction, everything harvesting food through to delivery to table, all of it involves some aspect of compute. And we soon will not comprehend nor create that underlying aspect of society. I dunno. It all seems like the answer to the fermi paradox.

spicyusername 3mo ago

Everyone is afraid about what will happen when people's skills atrophy, but the tools aren't going anywhere. If they truly can always recognize problems and communicate them in understandable ways, then people's skills won't atrophy. Everyone's skills will be enhanced. New developers will learn all of the things they need to learn through using the tool.

izzydata 3mo ago
The tools can definitely go away. They could get exponentially more expensive to where it is no longer financially possible to keep it subsidized. The bubble could pop and most of the companies could go bankrupt. It's not exactly like a calculator because it requires such an astronomical amount of resources to keep running.
- djmips 3mo ago
  I agree with you but there is a calculator analogy where the resources are local and while taxing for most, viable for the individual. Local AI is less capable than an army frontier model agents but still quite useful.
theodric 3mo ago
I think the most obvious impact (good AND bad) will be from people who didn't have (or want) skills to begin with being empowered to have a model crank out custom software to do little things to make their lives better. Can't atrophy a muscle that doesn't exist!

kobe_bryant 3mo ago

feel free to share even a single detail