Wait… this article… is AI-written?!? Lots of “It’s not X, it’s Y,” short choppy sentences, constantly re-summarizing itself and repeating itself… way longer than it needs to be… and I notice you’re not blaming the AI usage itself, just the inexperienced vibe coders.
I hate this. You realize you’re doing the same thing as the boxing company in this story? You’re hawking the work of professionals, in this case writers and editors, onto an AI that can’t actually replace them. It seems easy and magical, just like those warehouse workers thought with their shiny interface. A good writer would get to the point and not repeat themself. A real editor would cut this down to only what was needed.
I feel like I wasted my time reading this article that you didn’t even take the time to write. You should know better, especially with this wild and genuinely interesting story! Did it even happen, or did the AI hallucinate everything? What a hypocrite, you devalue writing AND IT SHOWS.
AI writing isn't the same as AI-assisted. I used to feel the same way as you. Explain how this is doing the same thing the boxing company is doing? One has you reading my article and the other can't fulfill orders. And this is me talking BTW. Or is it? How can you tell? Oh because pangram said so? I have an article for that too. Thanks for promoting my substack on yours. Maybe I did repeat too much. This just means we need more of your wife training these models. I appreciate your article as well.
“AI-assisted” is a bullshit term, like an executive taking writing credit for hiring someone to make a Batman story involving Darkseid.
I already did my comparison on why your AI usage is the same mindset as the boxing company. Has making Claude summarize things for you killed your reading comprehension?
I don’t use pangram, textual analysis is better. Your disdain for the craft disgusts me. You only care about “agentic engineers” and care fuck all for everyone else.
Here’s a link to my overall critique of this article for anyone interested:
Did I ask for literature? No, I asked for writing that gets to the point. Technical writing requires clarity and concision. And I don’t take reading advice from people who think reading and writing is a waste of time.
Nathan - Time to put your big boy pants on. You don't get to tell people how to write. I offered you a clear, concise, conciliatory way to control your baser instincts instead of just venting on the page. You obviously did not have a clue as to what I was saying. That is YOUR fault, not mine. AI is here to stay as long as the electrical grid still works. You don't have to like it. I don't either. But skimming the poor writing of AI-generated or AI-assisted writing is an easy fix. Get a grip.
I remember when I learned to program. The assembler language ran 10 pages, Binary ran 50, and Cobol ran 15+. I remember learning nested loops and integer+1.
This reminds me of the current problem with NASA and the old probes. Nobody codes Fortran anymore, and Cobol very rarely.
The thing with vibe coding is it’s supposed to be for vibes, not money. Anyone taking on clients should know enough to know when they’re over their head. That’s not an AI problem; it’s a business proficiency problem.
A good rule of thumb is not to vibe code anything you aren’t willing to pay the price for if it breaks.
People who should've known enough to know when they were in over their head but took clients on anyway existed long before generative AI. Before AI, the job was cleaning up human slop, now the job is cleaning up AI slop with AI assistance to wade through the additional LOC. Practically nothing has changed regarding custom SMB software.
Great article. It reminds me of the old story of the guy who walks into a broken server room, restarts a specific machine, and walks out with a $10,000 bill.
Absolutely. It's also already happening. https://youtu.be/LcLqC6M-cmo?si=rDNMsEpJGBHaRrxB and we're going to see a ton of knee jerk reactions. It'll be up to our group to clean up the mess from all the vibers. It's bad.
Once upon a time there was a role called a Business Analyst (or similar - check out IIBA) which used to do strange things things like speak to users in the business and then define requirements for a system that could support the business. They would work with the software designers and implementers and then do testing with the business.
Of course waterfall models are too expensive and take too long and who needs BAs anyway. So along came agile/scrum/kanban methodologies putting control in the hands if the developer.
Now AI vibe coding will need to learn all the lessons of the past.
I am SW methodology agnostic (use the right tool for the job). The message if this AI article seens to be to specify the requirements and design of the system correctly. Better give those BAs a call.
On a side note has anyone fed Claude a 500 page requirements document for some business system?
It was never meant to be profound. And that's ok. It's my views on what's happening. It's true that there used to be BAs. Then SEs ended up taking over those jobs because they knew the business and knew how to talk with customers. I'd say though that the SE had it harder than the BA unless of course the BAs were former SEs.
As for req docs. I got to 100 pages and Claude go through it and ask you a million questions. You'll end up in planning mode forever. 500 would be pure hell.
Nowadays, even before AI, the prototype is the core and the requirements work around it. It was bitch to go through all this req crap with the BA. Because after weeks after the mocks went up - everyone hated it - but it was in the requirements lol. Gotta love software eh?
THIS!!!! I've been trying to explain this to anyone who will listen. I'm not a developer but a retired IT manager who understands the business risk that all of this automated development is introducing. It used to be that businesses were risk adverse with mission critical systems. Systems and software that should and could not break didn't because going down would mean loss of profits and customers...and reputation. Now it seems that business is equating AI with "extreme money savings" which is leading to reckless business decisions because C-suite morons are looking all dreamy eyed when they hear "just fire high paid developers and replace them with easy to use cheap AI models...more profits will ensue." There is a reason that good developers are paid highly. There is great responsibility and knowledge involved with mission critical systems. Some idiot prompting Claude Code to get "something up and running as quickly and cheaply as possible" is so distant from measured, knowledgeable, and careful development.
They are learning that you get what you pay for...
This is exactly correct. Even if you tried to train the most sophisticated AI to drive an F1 car, the AI doesn't know the driving style of his teammates, the tendencies of his opponents, the feel of the car, etc. You either have real experienice or you get the painful experience. In both ways you end up paying for it. That's why the top people who get this are saying we need MORE of those experienced people - not less.
"But here’s the part a lot of engineers don’t want to hear. The basics won’t be enough.
These messes are massive. My box company’s codebase was generated in weeks, and multiply that by every rewrite prompt, every “optimize this,” every “fix the bug” layered on over eight months. You can’t figure that out by reading the code line by line. There’s too much of it, and it changes too fast. You’ll be untangling model-generated code with models. Pointing an agent at a codebase and asking it to trace where shipment state actually gets written. Asking it to map every place two processes can touch the same record. Asking it what happens when the carrier call succeeds and the database write doesn’t."
I ofton think it's error theory and the human condition that lead us here, I've always been an advocate of labor laws/job guarantees and others as these errors crop up. With supply scaling up and harnessing the inididual ability of the reverse curational relations. Or beaucraticing the neural nets.
Back to the main point for a minute. This story is not an example of "vibe coding hitting its limit". Its just an example of a company hiring lazy software developers (I use that term loosely) who chose to vibe code an enterprise level application. Who does that? Assuming that these guys could program conventionally, they would have probably produced exactly the same result. Wrong spec, game over. Its just another example of the fact that there's a lot more to software engineering than coding.
This is a very interesting content, but extremely hard to read because it’s probably AI written. Either that, or AI text is infecting people’s writing now.
please respect your readers time, no one wants to read gigantic blocks of LLM text
It would have garnered my upvote and I would follow you, were it not for the obvious fact that it is 93% AI written, at least according to the "Scan for AI text" provided by Substack in the upper right corner.
As others have observed, that's a real pity, and a bit of a shame. Call me censurious, if you wish.
Nitpicker would be better. ;) If AI corrected some of my grammer, does that make it a shame? Most of it was me. Seeing 93% is hilarious. Scan your comment and you’ll see that AI thought you wrote some percent of it. Lol.
Not really. I write a *lot* and at present with the same tool, I get 100% human-written pretty consistently, unless I'm quoting an AI text as part of my post.
But I appreciate that your style might just be AI-like---perhaps from reading too much AI text :)
Your article was nevertheless pretty darn good in terms of the important sentiments it conveyed.
This seems a lot more like bad assumptions upon implementation than vibe coding. I've encountered plenty of projects before AI that were a mess. I've had to talk to big clients and their engineering teams before that what they wanted to build and how they wanted to implement it wouldn't work. The people who knew the system the best didn't understand the flaws in their own approach.
Wait… this article… is AI-written?!? Lots of “It’s not X, it’s Y,” short choppy sentences, constantly re-summarizing itself and repeating itself… way longer than it needs to be… and I notice you’re not blaming the AI usage itself, just the inexperienced vibe coders.
I hate this. You realize you’re doing the same thing as the boxing company in this story? You’re hawking the work of professionals, in this case writers and editors, onto an AI that can’t actually replace them. It seems easy and magical, just like those warehouse workers thought with their shiny interface. A good writer would get to the point and not repeat themself. A real editor would cut this down to only what was needed.
I feel like I wasted my time reading this article that you didn’t even take the time to write. You should know better, especially with this wild and genuinely interesting story! Did it even happen, or did the AI hallucinate everything? What a hypocrite, you devalue writing AND IT SHOWS.
AI writing isn't the same as AI-assisted. I used to feel the same way as you. Explain how this is doing the same thing the boxing company is doing? One has you reading my article and the other can't fulfill orders. And this is me talking BTW. Or is it? How can you tell? Oh because pangram said so? I have an article for that too. Thanks for promoting my substack on yours. Maybe I did repeat too much. This just means we need more of your wife training these models. I appreciate your article as well.
“AI-assisted” is a bullshit term, like an executive taking writing credit for hiring someone to make a Batman story involving Darkseid.
I already did my comparison on why your AI usage is the same mindset as the boxing company. Has making Claude summarize things for you killed your reading comprehension?
I don’t use pangram, textual analysis is better. Your disdain for the craft disgusts me. You only care about “agentic engineers” and care fuck all for everyone else.
Here’s a link to my overall critique of this article for anyone interested:
https://nathanpdx.substack.com/p/its-not-x-its-y?r=1grn0m&utm_medium=ios&shareImageVariant=title
You just need to learn to skim. The essential idea is similar to Karl Popper's falsification concept. If you want literature, pick up Hemingway.
Did I ask for literature? No, I asked for writing that gets to the point. Technical writing requires clarity and concision. And I don’t take reading advice from people who think reading and writing is a waste of time.
Nathan - Time to put your big boy pants on. You don't get to tell people how to write. I offered you a clear, concise, conciliatory way to control your baser instincts instead of just venting on the page. You obviously did not have a clue as to what I was saying. That is YOUR fault, not mine. AI is here to stay as long as the electrical grid still works. You don't have to like it. I don't either. But skimming the poor writing of AI-generated or AI-assisted writing is an easy fix. Get a grip.
93% AI-assisted text.
I remember when I learned to program. The assembler language ran 10 pages, Binary ran 50, and Cobol ran 15+. I remember learning nested loops and integer+1.
This reminds me of the current problem with NASA and the old probes. Nobody codes Fortran anymore, and Cobol very rarely.
Remember GIGO?
I do.
Remember Basic on C64? That was me too
People do. A lot of climate modelling is done in Fortran.
The thing with vibe coding is it’s supposed to be for vibes, not money. Anyone taking on clients should know enough to know when they’re over their head. That’s not an AI problem; it’s a business proficiency problem.
A good rule of thumb is not to vibe code anything you aren’t willing to pay the price for if it breaks.
That's the key.
People who should've known enough to know when they were in over their head but took clients on anyway existed long before generative AI. Before AI, the job was cleaning up human slop, now the job is cleaning up AI slop with AI assistance to wade through the additional LOC. Practically nothing has changed regarding custom SMB software.
Great article. It reminds me of the old story of the guy who walks into a broken server room, restarts a specific machine, and walks out with a $10,000 bill.
“Wait he just pushed a button!”
“We're paying him to know which button to push.”
It also reminds me of this lyric video that lives rent free in my head now… https://youtu.be/3TNpOD6bov8?is=NXy6R8Yfq_4Y8VlU
We developers will be watching the headlines
Absolutely. It's also already happening. https://youtu.be/LcLqC6M-cmo?si=rDNMsEpJGBHaRrxB and we're going to see a ton of knee jerk reactions. It'll be up to our group to clean up the mess from all the vibers. It's bad.
Appreciate the article
Presentation issue: on stock iPhone, tables in article appear as very light grey on white, comments appear as dark grey on black
This article is not profound.
Once upon a time there was a role called a Business Analyst (or similar - check out IIBA) which used to do strange things things like speak to users in the business and then define requirements for a system that could support the business. They would work with the software designers and implementers and then do testing with the business.
Of course waterfall models are too expensive and take too long and who needs BAs anyway. So along came agile/scrum/kanban methodologies putting control in the hands if the developer.
Now AI vibe coding will need to learn all the lessons of the past.
I am SW methodology agnostic (use the right tool for the job). The message if this AI article seens to be to specify the requirements and design of the system correctly. Better give those BAs a call.
On a side note has anyone fed Claude a 500 page requirements document for some business system?
It was never meant to be profound. And that's ok. It's my views on what's happening. It's true that there used to be BAs. Then SEs ended up taking over those jobs because they knew the business and knew how to talk with customers. I'd say though that the SE had it harder than the BA unless of course the BAs were former SEs.
As for req docs. I got to 100 pages and Claude go through it and ask you a million questions. You'll end up in planning mode forever. 500 would be pure hell.
Nowadays, even before AI, the prototype is the core and the requirements work around it. It was bitch to go through all this req crap with the BA. Because after weeks after the mocks went up - everyone hated it - but it was in the requirements lol. Gotta love software eh?
THIS!!!! I've been trying to explain this to anyone who will listen. I'm not a developer but a retired IT manager who understands the business risk that all of this automated development is introducing. It used to be that businesses were risk adverse with mission critical systems. Systems and software that should and could not break didn't because going down would mean loss of profits and customers...and reputation. Now it seems that business is equating AI with "extreme money savings" which is leading to reckless business decisions because C-suite morons are looking all dreamy eyed when they hear "just fire high paid developers and replace them with easy to use cheap AI models...more profits will ensue." There is a reason that good developers are paid highly. There is great responsibility and knowledge involved with mission critical systems. Some idiot prompting Claude Code to get "something up and running as quickly and cheaply as possible" is so distant from measured, knowledgeable, and careful development.
They are learning that you get what you pay for...
This is exactly correct. Even if you tried to train the most sophisticated AI to drive an F1 car, the AI doesn't know the driving style of his teammates, the tendencies of his opponents, the feel of the car, etc. You either have real experienice or you get the painful experience. In both ways you end up paying for it. That's why the top people who get this are saying we need MORE of those experienced people - not less.
This all goes much better when you have run a team of juniors to produce actual working code that actually takes the edges into account…
You can get slop out of either but…
the robot types much faster so it can apply massive bandaids all over everything quickly while the humans talk about it :/
"But here’s the part a lot of engineers don’t want to hear. The basics won’t be enough.
These messes are massive. My box company’s codebase was generated in weeks, and multiply that by every rewrite prompt, every “optimize this,” every “fix the bug” layered on over eight months. You can’t figure that out by reading the code line by line. There’s too much of it, and it changes too fast. You’ll be untangling model-generated code with models. Pointing an agent at a codebase and asking it to trace where shipment state actually gets written. Asking it to map every place two processes can touch the same record. Asking it what happens when the carrier call succeeds and the database write doesn’t."
I ofton think it's error theory and the human condition that lead us here, I've always been an advocate of labor laws/job guarantees and others as these errors crop up. With supply scaling up and harnessing the inididual ability of the reverse curational relations. Or beaucraticing the neural nets.
Thx
Back to the main point for a minute. This story is not an example of "vibe coding hitting its limit". Its just an example of a company hiring lazy software developers (I use that term loosely) who chose to vibe code an enterprise level application. Who does that? Assuming that these guys could program conventionally, they would have probably produced exactly the same result. Wrong spec, game over. Its just another example of the fact that there's a lot more to software engineering than coding.
Well, AI does allow people to generate slop faster. So there is that.
Excellent Article! Build in haste, Repair in panic!
This is a very interesting content, but extremely hard to read because it’s probably AI written. Either that, or AI text is infecting people’s writing now.
please respect your readers time, no one wants to read gigantic blocks of LLM text
It is a technical article written by a developer for developers. Which “gigantic blocks of LLM text” makes it confusing to you?
Yeehaw!
Great article---spot on.
It would have garnered my upvote and I would follow you, were it not for the obvious fact that it is 93% AI written, at least according to the "Scan for AI text" provided by Substack in the upper right corner.
As others have observed, that's a real pity, and a bit of a shame. Call me censurious, if you wish.
Dr Jo.
Nitpicker would be better. ;) If AI corrected some of my grammer, does that make it a shame? Most of it was me. Seeing 93% is hilarious. Scan your comment and you’ll see that AI thought you wrote some percent of it. Lol.
Not really. I write a *lot* and at present with the same tool, I get 100% human-written pretty consistently, unless I'm quoting an AI text as part of my post.
But I appreciate that your style might just be AI-like---perhaps from reading too much AI text :)
Your article was nevertheless pretty darn good in terms of the important sentiments it conveyed.
This seems a lot more like bad assumptions upon implementation than vibe coding. I've encountered plenty of projects before AI that were a mess. I've had to talk to big clients and their engineering teams before that what they wanted to build and how they wanted to implement it wouldn't work. The people who knew the system the best didn't understand the flaws in their own approach.
Yea. Seems like a human engineer could make the exact same mistake if the people that knew the business didn’t give them correct instructions.