In case you yourself do not own a table saw, I should point out that table saws don't rip, cut, or mill stock for you. It's like saying "buying a hammer doesn't make you a carpenter".
In case anyone missed it, tptacek is saying that ai is different than just a tool. Tool analogies are not good.
A better analogy is, now you have what seems like an increasingly intelligent personal assistant that can do any cognitive work that you ask it to do, to increasingly better result, has retrograde amnesia, no accountability, entirely middle of the road morality, access to the internet... eh this isn't really a good analogy either.
Don't forget that when you're using it in your own domain you can at least catch its many mistakes and hallucinations, but "Citizen Devs" aren't going to have a knowledge base to do that.
I often wonder whether computer programmer said these kinds of things about spreadsheets, and how mere office workers would never have the knowledge base to build serious programs with them. Remember, before Visicalc, every spreadsheet was a program built by a computer programmer.
Nevertheless, those books aren't sold for the writing style, but because they're about the cult leader. So not the "writer" is prized (ghost or not) but the personality.
May I ask how you became credentialed in gatekeeping? Did you study gatekeeping in school, obtain a gatekeeping degree and pass the gatekeeping professional associations exam?
No? Then you are not qualified to say that other people are not qualified!
Whatever you can accomplish using AI without experience is now worthless. Sure that rando can now get a customized website hosted on Vercel with some struggling through setting up GitHub etc. That work is no longer interesting. It’s slop the same as a restaurant showing AI images of food. It’s digital plastic. AI on autopilot creating code, text, images or video is nothing more than a curiosity.
> AI on autopilot creating code, text, images or video is nothing more than a curiosity.
I wouldn’t say it’s lacking value. Maybe digital plastic is the perfect metaphor. Plastic is disposable and probably bad for the environment. But useful for all sorts of weird little tasks. Plastic is everywhere despite its weaknesses because sometimes you just want a thing shaped like this.
I’ve got a little personal data archive going back 20 years. But unfortunately it’s suffering bit rot. Last night I discovered the main hard drive storing it is suffering read failures - the zfs checksums aren’t matching. I pulled out Claude and told it to make a program that could make, maintain and verify a sidecar database of file hashes and reed-Solomon parity data so I can fix some small percent of bit rot that the drive experiences. A 10% increase in file size is worth it for the peace of mind of knowing that footage of my late grandma will still exist when I’m her age.
Is the program claude made isn’t art. I’m not putting it on my resume. If I had infinite time I’d love to make something like this lovingly by hand. But I’m working on other projects right now. This will be good enough. Maybe someone made something like this I could use instead. But this program is mine.
Hate to be that guy but wasnt the solution here to buy a new drive instead of trust claudes slop is long term maintaining your drive and not just looking like its doing a good job?
It's definitely an option. Though I'd really need 2 drives - one replacement, and a mirror. And then I could set up weekly sweeps which check zfs checksums to guard against bit rot. And just hope that bitrot never affects the same file on both drives at the same time.
I also don't want all my backups to be in the same physical location. If I copy some files with zfs to another drive, I'd also like to check that there wasn't an error in the copying process. Maybe I can use zfs to make mirror images - which probably copies the hashes?
Yeah, I could use zfs to do some of this. But having a sidecar checksum database is easier to work with across filesystems. And Reed-Solomon error correction is able to correct bit rot much more cheaply than using physically duplicated drives. I wouldn't mind buying a mirror set a year or two ago, but storage is so expensive now.
Filesystems in general aren't very good at protecting against silent bitrot, since it doesn't happen that much on spinning copper. But it happens a lot more than we'd all like to admit on modern SSDs. And zfs doesn't support Reed-Solomon error correction.
> In 28 may 2026, a 26 TB disk cost me 0.36 SEK per gigabyte.
The best price I can find for a 24gb+ hard drive is $1250 AUD for a 24gb Seagate Barracuda drive, at my local computer store. They're a similar price on amazon. Amazon US has refurbed drives too, which are a little cheaper but not much.
$1250 for 24gb is $52AUD per gigabyte, or 353 Swedish Krona per gigabyte.
Although AI substantially changes the situation the hard part of being an engineer was never the code.
A technician is defined by the tools they work with, an engineer is defined by the problems they can solve.
Taking a dozen different business processes and stakeholders and figuring out how to actually solve the problems they have is engineering.
There's lots of caveats asterisks & limitations, sure sure sure: we hear you Technically Correct repliers, here to spread your Well Actually naysayings.
But at the end of the day: heck yes. So much heck yes. Even if you are not engineering well, just the ability to have your agent go in and do reconissance of systems. "Reverse engineer this Bluetooth protocol for my device" or what I'm doing now, "reverse engineer my epaper display controller"... People have an incredibly powerful tool for assessing the world, for finding out.
And I for one strongly believe that will beget the great virous circle of lovely lovely fucking around. A wheel that is humanity's best invention, best motivation, best traction, best side, and one that any self proclaimed Hacker if any spirit ought be out of their seats cheering raving with excitement happy for.
My criteria is very simple, could you handwrite the code if you needed to. Could you write this code without the LLM writing it for you. Could you write this code without the IDE auto-completing it for you. You could take it all the way to, could you write the assembly if you needed to. Of course, the complexity of the ask and the timeline has to be commensurate with the power of your tooling at each level.
Do civil engineers need to be able to personally manufacture prestressed concrete? Or do they need to know the properties of materials and how to test their designs?
As a past civil engineer, I can't manufacture concrete.
But I still could calculate and write an electrical design project and draw all necessary plans and schemes without touching PC once – if I have a ruler, few pens and enough time (I would have to reference standards, but I did have most standards in print version)
I doubt your metaphor really fit. Was manufacturing ever an engineer job?
Today there are are specialties in the manufacture of prestressed concrete. Materials scientists, chemical engineers, metallurgical engineers, mechanical engineers, structural engineers, tooling engineers to design casting beds, quality control engineers, etc.
But in the early days of prestressed concrete (1920, 1930), a single civil engineer performed all of those functions.
The point is that the role of the modern civil engineer has changed and specialized over the decades. You can be a very competent civil engineer without having the knowledge or experience needed to perform every specialty.
Just because our ancestors in the 1940s handwrote binary sequences, and in the 1950s they handwrote assembly, and in the 1960s they handwrote FORTRAN, doesn't mean that today's software engineers need to be able to do any of that.
They don't need to because that's usually not what the job calls for, but I would think most people who call themselves software engineers believe they would be able to work at a job where they wrote FORTRAN instead of Python. It's just another programming language, you just have to spend a month to learn it. Assembly is just another programing language. Opcodes is just assembly without syntactic sugar. But of course that's not what my job calls for today, but if it did, it's not really any different in my mind from any new job where I'm doing something different, but just applying my software skills slightly differently.
I've noticed a near 1:1 overlap of people who advocate against shadow development, and people experiencing status anxiety about the fact that AI has proven that they barely pass the bar.
Some arguments about shadow development and IT are valid, like security and who takes ownership after the original developers leave, but they don't really account for the reality that organizations expand constantly, business needs are perpetually exploding and most development shops are so mired in business requirements and bureaucracy that there's genuinely no value involved with interacting with them.
Including software developers. Mathematicians tops unless the touch something outside code. An engine is literally multiple disciplines in one. I tired of SDs thinking they are engineers without touching hardware.
I've been using Claude to plan out some personal projects, and at first it seems like it can do anything!
But as I've used it, I've noticed that it overcomplicates everything and often even forgets what it previously said, and makes decisions that contradict it.
The AIs need a firm guiding hand, like a lead developer would give junior and mid-level programmers, and even occasionally to a senior developer that went off the rails for personal obsessions.
For instance, I was just trying to have it make a simple shopping list app. I had Claude take my old shopping info from the Google Keep list we use (it keeps the old checked-off items) and get the latest grocery ad, and produce a list of suggestions. It was amazing! I decided I want to send that info to a server, have a checklist there, and then keep the info for later so the AI could do it again.
After a lot of questioning (via Matt Pocock's skills, which are great) it was so complicated I finally yelled at it for overcomplicating everything, and told it how simple I wanted it to be, again.
It kept insisting on IDs for products, IDs for stores, IDs for store sections... Even after I told it no IDs. I finally said it a few more times in different ways.
It's finally getting back to the simple webapp that I imagined. It needed a lot of handholding to get it there.
And it's the kind of thing I would see from devs that were inflexible and had to see things a certain way.
I'm not ruling out that it could eventually get over this phase, but I have a feeling it'll always be something that needs handling.
My other project right now was naturally complex, and needed a lot of that stuff, and things have been going better, but I still correct it rather a lot, even on things I'd previously told it.
A vanishingly small percentage of the software in the world has failure modes that kill people. To say the analogy you're making is "only partly applicable" would be very generous..
No, but a shocking amount of unappreciated software can cost a TON of money if it fails, even briefly, which is another disaster scenario.
More software than you'd think can get people killed, though.
And more software than you'd think is running inside a barely-understood SCADA package that runs user scripts, which can now be authored by AI the user doesn't understand, and that can result in either or both of the above outcomes. This is true in just about every city in the world, several times over.
Software engineers have to understand that they aren't the only flavor of engineer, and even if "everyone is a [software] engineer" was the intended statement, it demonstrates excessive hubris. Many real engineers do work on systems with failure modes that kill people.
Good. IT broadly and unnecessarily adopted the complexity of mega-sites that serve hundreds of millions of users even when their actual user base was in the hundreds. AI allows competent and detail-oriented people with technical knowledge to again create useful internal tools in hours or day instead of the "never" timeline from gatekeeping developers.
I don't want to be rude, but if you're retired by now you should stay retired.
Your comment fits a pattern I see on HN all the time from the older crowd that was promoted to management way too early in their career. It's not helping.
Found this post super interesting. I've personally run into similar, and built a really lightweight k8s operator and claude skill that lets the citizen devs deploy to an eng/IT managed internal k8s cluster. Keeps things safe (non-public), and can update it as they get more advanced.
I was very anti it then - it was clearly Consultancy BS; but now non-devs really do have much more scope to create things that were only possible by software engineers before - and yes that creates challenges for us professionals.
But I no longer sneer at "The Citizen Developer" as an idea; its time has come.
New person in sales joined my company a few months ago and on the very first introduction they handed me a vibe coded app and asked me to review it for them and put it into production. Haven't spoken to them since. They haven't proposed anything else since then either to engineering.
You may think it harsh but they didn't even ask my name, where I work, what I do etc.
Engineering to these folks is a one-and-done endeavour. You just prompt and fire it off. They never see the work that goes on after the launch party. The midnight debugging sessions, the database migrations, the debates over which auth provider to use, the list goes on. Until the "citizen developer" is willing to engage in the less "sexy" part of software engineering, the world will move on without them as well.
62 comments
[ 0.70 ms ] story [ 7.1 ms ] threadThe level of propaganda reaches Maoist proportions.
https://news.ycombinator.com/newsguidelines.html
I think it's pretty obvious what parent poster was implying. Pedantic nitpicks add nothing to the discussion.
A better analogy is, now you have what seems like an increasingly intelligent personal assistant that can do any cognitive work that you ask it to do, to increasingly better result, has retrograde amnesia, no accountability, entirely middle of the road morality, access to the internet... eh this isn't really a good analogy either.
They actually do all of that
That usually requires a 3 to 5 year degree, and possibly professional exam.
Having the AI do the work doesn't make one an engineer.
Just like one doesn't become a writer by owning a typewriter or a word processor.
Isn’t this what society already largely rewards? See the personality cults around CEOs and not even founders.
No? Then you are not qualified to say that other people are not qualified!
Do you see the absurdity of your position now?
I wouldn’t say it’s lacking value. Maybe digital plastic is the perfect metaphor. Plastic is disposable and probably bad for the environment. But useful for all sorts of weird little tasks. Plastic is everywhere despite its weaknesses because sometimes you just want a thing shaped like this.
I’ve got a little personal data archive going back 20 years. But unfortunately it’s suffering bit rot. Last night I discovered the main hard drive storing it is suffering read failures - the zfs checksums aren’t matching. I pulled out Claude and told it to make a program that could make, maintain and verify a sidecar database of file hashes and reed-Solomon parity data so I can fix some small percent of bit rot that the drive experiences. A 10% increase in file size is worth it for the peace of mind of knowing that footage of my late grandma will still exist when I’m her age.
Is the program claude made isn’t art. I’m not putting it on my resume. If I had infinite time I’d love to make something like this lovingly by hand. But I’m working on other projects right now. This will be good enough. Maybe someone made something like this I could use instead. But this program is mine.
Is it slop? Yes. Is it useful? Yes.
I also don't want all my backups to be in the same physical location. If I copy some files with zfs to another drive, I'd also like to check that there wasn't an error in the copying process. Maybe I can use zfs to make mirror images - which probably copies the hashes?
Yeah, I could use zfs to do some of this. But having a sidecar checksum database is easier to work with across filesystems. And Reed-Solomon error correction is able to correct bit rot much more cheaply than using physically duplicated drives. I wouldn't mind buying a mirror set a year or two ago, but storage is so expensive now.
Filesystems in general aren't very good at protecting against silent bitrot, since it doesn't happen that much on spinning copper. But it happens a lot more than we'd all like to admit on modern SSDs. And zfs doesn't support Reed-Solomon error correction.
Yea, with just a single drive it's not really a backup is it?
> If I copy some files with zfs to another drive, I'd also like to check that there wasn't an error in the copying process.
That's what rsync is for.
The best price I can find for a 24gb+ hard drive is $1250 AUD for a 24gb Seagate Barracuda drive, at my local computer store. They're a similar price on amazon. Amazon US has refurbed drives too, which are a little cheaper but not much.
$1250 for 24gb is $52AUD per gigabyte, or 353 Swedish Krona per gigabyte.
Visual Basic
Low-Code
Power Platform
Although AI substantially changes the situation the hard part of being an engineer was never the code.
A technician is defined by the tools they work with, an engineer is defined by the problems they can solve. Taking a dozen different business processes and stakeholders and figuring out how to actually solve the problems they have is engineering.
There's lots of caveats asterisks & limitations, sure sure sure: we hear you Technically Correct repliers, here to spread your Well Actually naysayings.
But at the end of the day: heck yes. So much heck yes. Even if you are not engineering well, just the ability to have your agent go in and do reconissance of systems. "Reverse engineer this Bluetooth protocol for my device" or what I'm doing now, "reverse engineer my epaper display controller"... People have an incredibly powerful tool for assessing the world, for finding out.
And I for one strongly believe that will beget the great virous circle of lovely lovely fucking around. A wheel that is humanity's best invention, best motivation, best traction, best side, and one that any self proclaimed Hacker if any spirit ought be out of their seats cheering raving with excitement happy for.
But I still could calculate and write an electrical design project and draw all necessary plans and schemes without touching PC once – if I have a ruler, few pens and enough time (I would have to reference standards, but I did have most standards in print version)
I doubt your metaphor really fit. Was manufacturing ever an engineer job?
But in the early days of prestressed concrete (1920, 1930), a single civil engineer performed all of those functions.
The point is that the role of the modern civil engineer has changed and specialized over the decades. You can be a very competent civil engineer without having the knowledge or experience needed to perform every specialty.
Just because our ancestors in the 1940s handwrote binary sequences, and in the 1950s they handwrote assembly, and in the 1960s they handwrote FORTRAN, doesn't mean that today's software engineers need to be able to do any of that.
Some arguments about shadow development and IT are valid, like security and who takes ownership after the original developers leave, but they don't really account for the reality that organizations expand constantly, business needs are perpetually exploding and most development shops are so mired in business requirements and bureaucracy that there's genuinely no value involved with interacting with them.
> business needs are perpetually exploding
Ignorance is perpetually overthinking. If you want to be an engineer, you should quit your current role and reapply instead of ranting on here.
But as I've used it, I've noticed that it overcomplicates everything and often even forgets what it previously said, and makes decisions that contradict it.
The AIs need a firm guiding hand, like a lead developer would give junior and mid-level programmers, and even occasionally to a senior developer that went off the rails for personal obsessions.
For instance, I was just trying to have it make a simple shopping list app. I had Claude take my old shopping info from the Google Keep list we use (it keeps the old checked-off items) and get the latest grocery ad, and produce a list of suggestions. It was amazing! I decided I want to send that info to a server, have a checklist there, and then keep the info for later so the AI could do it again.
After a lot of questioning (via Matt Pocock's skills, which are great) it was so complicated I finally yelled at it for overcomplicating everything, and told it how simple I wanted it to be, again.
It kept insisting on IDs for products, IDs for stores, IDs for store sections... Even after I told it no IDs. I finally said it a few more times in different ways.
It's finally getting back to the simple webapp that I imagined. It needed a lot of handholding to get it there.
And it's the kind of thing I would see from devs that were inflexible and had to see things a certain way.
I'm not ruling out that it could eventually get over this phase, but I have a feeling it'll always be something that needs handling.
My other project right now was naturally complex, and needed a lot of that stuff, and things have been going better, but I still correct it rather a lot, even on things I'd previously told it.
More software than you'd think can get people killed, though.
And more software than you'd think is running inside a barely-understood SCADA package that runs user scripts, which can now be authored by AI the user doesn't understand, and that can result in either or both of the above outcomes. This is true in just about every city in the world, several times over.
Your comment fits a pattern I see on HN all the time from the older crowd that was promoted to management way too early in their career. It's not helping.
https://news.ycombinator.com/item?id=22034221
I was very anti it then - it was clearly Consultancy BS; but now non-devs really do have much more scope to create things that were only possible by software engineers before - and yes that creates challenges for us professionals.
But I no longer sneer at "The Citizen Developer" as an idea; its time has come.
Engineering to these folks is a one-and-done endeavour. You just prompt and fire it off. They never see the work that goes on after the launch party. The midnight debugging sessions, the database migrations, the debates over which auth provider to use, the list goes on. Until the "citizen developer" is willing to engage in the less "sexy" part of software engineering, the world will move on without them as well.