2 ms·

I know this post is about tech, and obviously HN is a tech centered site, and most people here work in tech, but let me say: Outside of tech, AI has been pheno

I know this post is about tech, and obviously HN is a tech centered site, and most people here work in tech, but let me say: Outside of tech, AI has been phenomenally helpful. I know many tech folk are falling over themselves for non-tech industry problems that can be software-solved then leased out monthly, and there are tons of these problems out there, but very hard to locate and model if you are outside the industry. But with the current crop of LLMs, people who don't know how to program, but recognize that a program could do this task, finally can now summon that program to do the task. The path still has a tech-ability moat, but I can only imagine the AI titans racing to get programming ability into Supply Chain Technician Kim's hands. Think Steve Jobs designing an IDE for your mother to use. I believe it will be the CEO's of these non-tech companies that will be pushing "AI first" and having people come in to show non-techy non-tech workers how to leverage LLMs to automate tasks. You guys have to keep in mind that if you walk into most offices in most places of the world, most workers will say "What the hell is a macro? I just go down the list line by line..."

dowager_dan99 2y ago
>> But with the current crop of LLMs, people who don't know how to program, but recognize that a program could do this task, finally can now summon that program to do the task. The path still has a tech-ability moat, but I can only imagine the AI titans racing to get programming ability into Supply Chain Technician Kim's hands. Think Steve Jobs designing an IDE for your mother to use Brooks has yet to be proven wrong; even if this appears to be the silver bullet it could be just as likely that this widens the tech moat when non-programmers paint themselves into corners where they can't do their jobs without all the brittle, impossible to maintain code they've written. Think of the skilled trades vacuum we have in much of the Western world. Can Supply Chain Technician Kim Jr do her job without AI if she's never seen that before?
- Workaccount2 2y ago
  I think the core disconnect is that the current tech paradigm is writing broad scope feature packed programs (and it kind of needs to be that way). One stop shops for every need that the broadest possible customer base could have, while minimizing time that needs to be spent on support. But I don't see that being the future. Instead I think people will just spin up bespoke ultra narrow scope programs (maybe scripts is more fitting here, but people like GUIs - scripts with GUIs?) that are generally under 3K LOC. You don't need an AI to one-shot Excel.exe if you just want a simple way to track how many plastic pellets came in today, and how many went out. A GUI on a simple program on an SQlite database will do that no problem. And you can ditch that bloated excel doc you have been using for years. At my own company we forwent a proprietary CAD package because Claude could decode the files we had, make a GUI for doing the transformation we needed to do, and properly reencode the file.
remich 2y ago
Re your third paragraph - can they though? I'm not trying to be combative, it's just that what you wrote has been the promise of all no-code/low-code technologies for literal decades. The exclamation "finally, all of the back-office people can write their own software!" is just the other side of the "finally, we can get rid of all the software engineers!" coin. But, so far, every single other time this has been tried it runs into the ease-of-use / customizability problem. The easier a tool is to use/learn, the harder it is (or impossible) to use for specific use-cases. And, vice versa, the more flexible / customizable a tool is, the harder it is to use/learn (looking at you Jira). Maybe this time is actually different, but I'll believe it when I see it.
- ctkhn 2y ago
  Yep - the easier it is to go way down a hole the harder it is to dig yourself out.

dowager_dan99 2y ago

>> But with the current crop of LLMs, people who don't know how to program, but recognize that a program could do this task, finally can now summon that program to do the task. The path still has a tech-ability moat, but I can only imagine the AI titans racing to get programming ability into Supply Chain Technician Kim's hands. Think Steve Jobs designing an IDE for your mother to use Brooks has yet to be proven wrong; even if this appears to be the silver bullet it could be just as likely that this widens the tech moat when non-programmers paint themselves into corners where they can't do their jobs without all the brittle, impossible to maintain code they've written. Think of the skilled trades vacuum we have in much of the Western world. Can Supply Chain Technician Kim Jr do her job without AI if she's never seen that before?

Workaccount2 2y ago
I think the core disconnect is that the current tech paradigm is writing broad scope feature packed programs (and it kind of needs to be that way). One stop shops for every need that the broadest possible customer base could have, while minimizing time that needs to be spent on support. But I don't see that being the future. Instead I think people will just spin up bespoke ultra narrow scope programs (maybe scripts is more fitting here, but people like GUIs - scripts with GUIs?) that are generally under 3K LOC. You don't need an AI to one-shot Excel.exe if you just want a simple way to track how many plastic pellets came in today, and how many went out. A GUI on a simple program on an SQlite database will do that no problem. And you can ditch that bloated excel doc you have been using for years. At my own company we forwent a proprietary CAD package because Claude could decode the files we had, make a GUI for doing the transformation we needed to do, and properly reencode the file.

remich 2y ago

Re your third paragraph - can they though? I'm not trying to be combative, it's just that what you wrote has been the promise of all no-code/low-code technologies for literal decades. The exclamation "finally, all of the back-office people can write their own software!" is just the other side of the "finally, we can get rid of all the software engineers!" coin. But, so far, every single other time this has been tried it runs into the ease-of-use / customizability problem. The easier a tool is to use/learn, the harder it is (or impossible) to use for specific use-cases. And, vice versa, the more flexible / customizable a tool is, the harder it is to use/learn (looking at you Jira). Maybe this time is actually different, but I'll believe it when I see it.

ctkhn 2y ago
Yep - the easier it is to go way down a hole the harder it is to dig yourself out.