I don’t disagree. Most generative AI models are some variant on “plagiarism machine”, but categorizing and identifying data are extremely useful things that AI does.
LLMs are good at quickly generating code, but the issue in software is rarely how fast humans can write code. In fact, more speed with less understanding is a really bad combination (I am a developer working DevOps and anecdotally I see way more large scale bugs now than I did 5 years ago).
Agentic AI is, unfortunately, just an LLM pretending to be a person, and that’s a really bad thing. Like so incredibly bad. Did you know that humans are statistically more likely to make mistakes when under pressure? Cause the LLMs sure do. Create a narrative of pressure and the LLM cracks like a rotten egg. Cause that’s more statistically likely!
You’ve accurately described why having LLMs cobble together code is a terrible idea. With all the vibe-coded nonsense finding its way into production code because the amount of code generated in a short amount of time inevitably overwhelms human oversight, I wouldn’t want to work in cybersec these days.
That said, I do see applications for LLMs in areas where mistakes will not get people hurt or systems breached: they can
provide a first layer of customer and tech support to solve the really stupid stuff that needs no human attention (“Have you made sure that the device is plugged in? Have you tried turning it off and on again?”). This can be particularly useful when paired with a source of truth it can draw from.
do tedious tasks involving large amounts of text processing, e.g. automated translation, cross-referencing of legal texts.
provide pre-college learners with a tutor that is available 24/7, can explain simple academic subject matters and answer questions that naturally arise as part of every learning process. Again, pair with a source of truth for more reliable results.
The speed and ease at which LLMs allow you to generate code is a bug, not a feature in my opinion. In my org, a group of 3 very junior engineers wrote a 5k line shell script for building k8s clusters according to our business specs and it’s fucking awful. The actual time to get it out the door was short, but now it’s basically impossible to change it without fucking up like 20 different things. The fucking thing will randomly quit because the shit ass LLM thinks set -e is a good thing to use, and it’s full of unused variables everywhere. I had to add a feature to it (which is how I learned of its existence), and I spent a miserable week just reading the entire fucking thing so I could ensure that my change wouldn’t cause an oil refinery in the North Sea to explode due to a butterfly-effect series of bullshit.
The frustration and toil you feel as a software dev is a feature. If something is making you mad and is taking forever to write, that’s a sign you probably need to change your approach. If you’re using an LLM to write a bunch of boilerplate, why not just eliminate the boilerplate or like, make a factory to spit out a bunch of it or something? Your discomfort is a powerful tool and you are not best served by ignoring it. Those junior devs would have written something much better if they had been forced to experience the true toil and suffering of writing a 5k line shell script.
I fall in the junior category, but with more experience prior to becoming a certified full stack dev than most juniors. I was a sys admin for a decade where I taught myself how to code to simplify my job. Plus I had 1 year at university 15 years ago. I use my company provided license with AI very sparingly and never let it implement code. Mostly I use it like a glorified stack overflow when I run into a problem that I can’t work out by myself. Usually, it will suggest some code that’s not good, but it’s enough to highlight a concept I’m ignorant to and then I can do what I need. If there’s a block of code that has something that I don’t understand, I can highlight it and ask it to explain. It’s usually pretty good at listing out what something is doing or at least supposed to do.
I would love if AI disappeared immediately, but it’s not going to happen. If someone is using, it should be used as a tool and not a replacement. If you can’t do the thing that it’s doing, then you shouldn’t be use it to do that thing. I probably ask it questions less than once a week, and again, never put in code that I don’t understand what it does and why.
I have a good friend that’s a senior dev at a company using Claude code. He’s become an AI code reviewer, but much to my dismay likes it. He’s vibe coding his own fuck around app with it and it’s writing the backend in C#, a language he doesn’t know being a frontend dev. It’s so infuriating to me that someone that I know to be intelligent is so damn stupid.
I’ll say that given the way OpenAI and Anthropic have hideously overextended themselves (they have over a trillion dollars of financial commitments to companies like Oracle), it’s not impossible that the current crop of American LLM providers do just kinda… poof away. Traditional banks want nothing more to do with them, they’re getting majorly spooked. All that needs to happen is for private credit to lose confidence in them, which is already happening. When they’re out of cash, they’ll be on the hook for an absolute ridiculous amount of money and they’ll probably just get liquidated.
I’m sure there will be new companies that pop up, but they’re going to have to charge 10x what Anthropic is making enterprise customers pay, since inference likely still isn’t profitable at the price Anthropic is charging.
I don’t use LLMs as a knowledge base because if the problem is bad enough for me, I’m likely just grepping through kubernetes source code or something. That being said, I don’t necessarily have an issue with folks using an LLM that way as long as they fully understand exactly how bad it is at what it does. You’ll be fine if you lose access to LLMs, and that’s the number one thing in my book. Your friend? Not so much.
(As a fun bonus to all of this, Oracle is very likely to die if OpenAI can’t meet its commitments. Either way, Larry Ellison will probably stop being a billionaire, since almost all of his wealth is in Oracle stock).
The schadenfreude I will experience if I see these giants fall will be immeasurable. I’ll have to go visit my doctor have 4 hours, for reasons. I would love to see oracle, openAI, and anthropomorphic just dissolve, but it would be extra special if Microsoft was broken down to nothing. Google too, but while I don’t like Google, they’re definitely lower on the list than those first options, maybe ahead of Oracle, but mostly because I’m fortunate enough to not have to use any of their products.
Umm, excuse me. We’re delivering 400 points worth of story work with 40 points worth of dev time. Do you think that that’s somehow a bad thing? Our budget is stretching further than ever! (Once we figure out how to reduce token costs) /s
The issue is more that the triangle of code that works, code that scales well, and code that’s cheap will always, ALWAYS, prioritize works and is cheap, even if every action taken from then on costs more to make. I’ve been on a team that focused effort on keeping scalability a priority and every single thing we tried got kneecapped to “keep to the budget”.
I don’t disagree. Most generative AI models are some variant on “plagiarism machine”, but categorizing and identifying data are extremely useful things that AI does.
LLMs are good at quickly generating code, but the issue in software is rarely how fast humans can write code. In fact, more speed with less understanding is a really bad combination (I am a developer working DevOps and anecdotally I see way more large scale bugs now than I did 5 years ago).
Agentic AI is, unfortunately, just an LLM pretending to be a person, and that’s a really bad thing. Like so incredibly bad. Did you know that humans are statistically more likely to make mistakes when under pressure? Cause the LLMs sure do. Create a narrative of pressure and the LLM cracks like a rotten egg. Cause that’s more statistically likely!
You’ve accurately described why having LLMs cobble together code is a terrible idea. With all the vibe-coded nonsense finding its way into production code because the amount of code generated in a short amount of time inevitably overwhelms human oversight, I wouldn’t want to work in cybersec these days.
That said, I do see applications for LLMs in areas where mistakes will not get people hurt or systems breached: they can
The speed and ease at which LLMs allow you to generate code is a bug, not a feature in my opinion. In my org, a group of 3 very junior engineers wrote a 5k line shell script for building k8s clusters according to our business specs and it’s fucking awful. The actual time to get it out the door was short, but now it’s basically impossible to change it without fucking up like 20 different things. The fucking thing will randomly quit because the shit ass LLM thinks
set -eis a good thing to use, and it’s full of unused variables everywhere. I had to add a feature to it (which is how I learned of its existence), and I spent a miserable week just reading the entire fucking thing so I could ensure that my change wouldn’t cause an oil refinery in the North Sea to explode due to a butterfly-effect series of bullshit.The frustration and toil you feel as a software dev is a feature. If something is making you mad and is taking forever to write, that’s a sign you probably need to change your approach. If you’re using an LLM to write a bunch of boilerplate, why not just eliminate the boilerplate or like, make a factory to spit out a bunch of it or something? Your discomfort is a powerful tool and you are not best served by ignoring it. Those junior devs would have written something much better if they had been forced to experience the true toil and suffering of writing a 5k line shell script.
I fall in the junior category, but with more experience prior to becoming a certified full stack dev than most juniors. I was a sys admin for a decade where I taught myself how to code to simplify my job. Plus I had 1 year at university 15 years ago. I use my company provided license with AI very sparingly and never let it implement code. Mostly I use it like a glorified stack overflow when I run into a problem that I can’t work out by myself. Usually, it will suggest some code that’s not good, but it’s enough to highlight a concept I’m ignorant to and then I can do what I need. If there’s a block of code that has something that I don’t understand, I can highlight it and ask it to explain. It’s usually pretty good at listing out what something is doing or at least supposed to do.
I would love if AI disappeared immediately, but it’s not going to happen. If someone is using, it should be used as a tool and not a replacement. If you can’t do the thing that it’s doing, then you shouldn’t be use it to do that thing. I probably ask it questions less than once a week, and again, never put in code that I don’t understand what it does and why.
I have a good friend that’s a senior dev at a company using Claude code. He’s become an AI code reviewer, but much to my dismay likes it. He’s vibe coding his own fuck around app with it and it’s writing the backend in C#, a language he doesn’t know being a frontend dev. It’s so infuriating to me that someone that I know to be intelligent is so damn stupid.
I’ll say that given the way OpenAI and Anthropic have hideously overextended themselves (they have over a trillion dollars of financial commitments to companies like Oracle), it’s not impossible that the current crop of American LLM providers do just kinda… poof away. Traditional banks want nothing more to do with them, they’re getting majorly spooked. All that needs to happen is for private credit to lose confidence in them, which is already happening. When they’re out of cash, they’ll be on the hook for an absolute ridiculous amount of money and they’ll probably just get liquidated.
I’m sure there will be new companies that pop up, but they’re going to have to charge 10x what Anthropic is making enterprise customers pay, since inference likely still isn’t profitable at the price Anthropic is charging.
I don’t use LLMs as a knowledge base because if the problem is bad enough for me, I’m likely just grepping through kubernetes source code or something. That being said, I don’t necessarily have an issue with folks using an LLM that way as long as they fully understand exactly how bad it is at what it does. You’ll be fine if you lose access to LLMs, and that’s the number one thing in my book. Your friend? Not so much.
(As a fun bonus to all of this, Oracle is very likely to die if OpenAI can’t meet its commitments. Either way, Larry Ellison will probably stop being a billionaire, since almost all of his wealth is in Oracle stock).
The schadenfreude I will experience if I see these giants fall will be immeasurable. I’ll have to go visit my doctor have 4 hours, for reasons. I would love to see oracle, openAI, and anthropomorphic just dissolve, but it would be extra special if Microsoft was broken down to nothing. Google too, but while I don’t like Google, they’re definitely lower on the list than those first options, maybe ahead of Oracle, but mostly because I’m fortunate enough to not have to use any of their products.
Umm, excuse me. We’re delivering 400 points worth of story work with 40 points worth of dev time. Do you think that that’s somehow a bad thing? Our budget is stretching further than ever! (Once we figure out how to reduce token costs) /s
The issue is more that the triangle of code that works, code that scales well, and code that’s cheap will always, ALWAYS, prioritize works and is cheap, even if every action taken from then on costs more to make. I’ve been on a team that focused effort on keeping scalability a priority and every single thing we tried got kneecapped to “keep to the budget”.