A great blog post that applied Occam's Razor to AI coding claims, and then backed that up with some data. The premise is simple: If AI programming assistance was really increasing productivity as...
A great blog post that applied Occam's Razor to AI coding claims, and then backed that up with some data.
The premise is simple: If AI programming assistance was really increasing productivity as much as reported, there would be a gargantuan increase in toy projects and a flooding of all app stores with garbage since 2020.
And by all accounts, that hasn't happened. And it adds a tragic cap to the rounds of layoffs occuring because of the bullshit productivity claims.
It's stupid to fire engineers for not using AI coding tools, but I think that's basically post-hoc justification. In general, the C suite is looking for an excuse to do layoffs. If revenue isn't...
It's stupid to fire engineers for not using AI coding tools, but I think that's basically post-hoc justification.
In general, the C suite is looking for an excuse to do layoffs. If revenue isn't growing, the only way to increase profit is cutting costs. When a company does layoffs, they can be honest and say they're cutting headcount and likely losing productive capacity, or they can lie about massive productivity growth through AI tooling.
It isn't much different from doing layoffs through RTO policies. The evidence on productivity isn't there, but executives will gladly lie through their teeth to shirk the blame.
Or they can say nothing, because they're not obligated to give a reason. Which is what they tend to do in pretty much every case. Most of these "X is doing layoffs because of Y" headlines are...
When a company does layoffs, they can be honest and say they're cutting headcount and likely losing productive capacity, or they can lie about massive productivity growth through AI tooling.
Or they can say nothing, because they're not obligated to give a reason. Which is what they tend to do in pretty much every case.
Most of these "X is doing layoffs because of Y" headlines are heavily mined quotes from interviews of people who may not even be at a leadership level where they had any control over it very much after the fact.
I feel like I see a lot of random anthropomorphizing of companies. Price increases and layoffs are a common one. Companies don't need to, nor do they, talk to anyone when they do either. They don't need an excuse.
They sure are obligated to give a reason to their shareholders and they almost always do. The topic frequently comes up in earnings calls. It just so happens that efficiency gains as a reason...
They sure are obligated to give a reason to their shareholders and they almost always do. The topic frequently comes up in earnings calls. It just so happens that efficiency gains as a reason shows a lot more certainty than layoffs do.
Even moreso in Europe. They can't fire me without cause and must have a good reason to lay me off that needs to be signed off by a work council. In fact I was recently laid off and was first told their plans for the future and the reason why they were "making the organisation leaner" before they said the actual words that I was fired.
I know that my US counterparts were usually told less and escorted out of the door the minute after the meeting, something absolutely not allowed here, but even the US org needed to provide a reason to their shareholders.
There are two sides to the labor coin. The cost, and the productivity. Shareholders love hearing about the cost part: "we slashed our operating costs by 60% by firing 1000 developers!!!" They...
There are two sides to the labor coin. The cost, and the productivity. Shareholders love hearing about the cost part: "we slashed our operating costs by 60% by firing 1000 developers!!!" They don't like to hear about the productivity part: "our new feature releases will now take 5 times as long".
AI lets them pretend that they get to do the first part without the second part.
While that’s very true in a basic sense, I’d wager the hype-like promise of “10x productivity must mean 10x revenue soon!!” would sell better, since it doesn’t imply stagnant revenues, so that is...
While that’s very true in a basic sense, I’d wager the hype-like promise of “10x productivity must mean 10x revenue soon!!” would sell better, since it doesn’t imply stagnant revenues, so that is probably what firms prefer for outward communication.
A little while ago this was posted on tildes: No, AI is not making engineers 10x as productive: curing your AI 10x engineer imposter syndrome I think it complements this article quite nicely by...
That really nails it for me. The hype has really made it impossible to discuss how these types of programs are best used. It’s all focused on these fantastical, future claims of massive social...
while I do get use out of LLMs most of my use cases are underwhelming compared to the hype.
That really nails it for me. The hype has really made it impossible to discuss how these types of programs are best used. It’s all focused on these fantastical, future claims of massive social change.
It does seem like normal, non-tech people are starting to get annoyed at having AI shoved in every single product. So hopefully we’ve reached the crest of the wave and it can come crashing down.
AI is in a weird place for me as a developer. It does sometimes make certain aspects of my work faster, in the moment, but usually it also means I miss out on a whole research/discovery process...
AI is in a weird place for me as a developer. It does sometimes make certain aspects of my work faster, in the moment, but usually it also means I miss out on a whole research/discovery process that massively increases my knowledge of a subject, which in turn means other work now becomes slower because I didn't learn as much.
Partly because of that, and because I tend to not trust the AI, I end up using it as a suggestion towards a solution and then look to stackoverflow or the docs to back it up, and make sure i understand it all. That's not really efficient.
Its good for a few things Helps with best practices in a new language (ex how to do blah general task in Go) Helps newbies learn quicker by generating specific answers without having to wait for...
Its good for a few things
Helps with best practices in a new language (ex how to do blah general task in Go)
Helps newbies learn quicker by generating specific answers without having to wait for someone to respond (why does python use if name == main? With free followup questions
Builds good templates for things like yml files
Does a pretty good job of reading documentation for you, but when its wrong, boy is it wrong.
I don’t think a developer with over 20 years experience is using it much for these things, the things that its good at.
Idk what these senior guys are using it for, really, cause we all know it doesn’t work great within complex codebases, too much needed context, and Im convinced anyone who says that it can do that is lying.
Then you would be wrong. I have over 25 years of experience and these days I use Claude Code almost full-time. That is because I am currently working in languages that I don't really know, but...
I don’t think a developer with over 20 years experience is using it much for these things, the things that its good at.
Then you would be wrong. I have over 25 years of experience and these days I use Claude Code almost full-time.
That is because I am currently working in languages that I don't really know, but that are related to languages/technologies that I do know. Last week I spent two days setting up a (for me) complicated cloud configuration on GCP with Terraform. I have experience with Pulumi, I have experience with configuring those types of environments on GCP and AWS, but I had actually never used Terraform before. I can explain in clear terms what I want from Claude and I know which questions to ask when I read the code it spits out. In addition it turns out Claude is incredibly fast at writing gcloud commands that help debug something that isn't quite right. Really, if I had had to learn all that from the documentation, it would have taken me days extra, if not a week.
The weeks before that I was writing web services in Python. Same thing. I have over two decades of experience in writing internet backends, but somehow I have never done one in Python. I know how to ask for what I want and I know how to test and debug what I get back. Of course, I could have written it all by myself, and it would not have taken me months or weeks. But simply not having to worry about the exact syntax, or figuring out what the standard libraries for X or Y are, is a massive timesaver.
I think plenty of seniors end up on a wide variety of projects. I've seen seniors with too much bluster try to learn a front-end framework in a weekend...
I think plenty of seniors end up on a wide variety of projects. I've seen seniors with too much bluster try to learn a front-end framework in a weekend...
I'm not sure how to respond to your last sentence, but I'll take the challenge. I'm not an AI priponent, but I use it at work and it definitely makes me faster. I have twenty five years as a...
I'm not sure how to respond to your last sentence, but I'll take the challenge. I'm not an AI priponent, but I use it at work and it definitely makes me faster. I have twenty five years as a software engineer.
My codebase is a well organized monorepo of related services and tools. They're joined together with an open API spec, strict typescript types, and a DB schema. I mention those because it think AI does best with strong metadata about the code, which these things are. I also have a lot of unit tests.
If I ask for an update, I usually describe it in terms of a DB schema update and a corresponding API change. AI usually does a good job of solving for me. It updates the spec and schema first, then the rest flows nicely. It makes my changes, fixes typing errors, runs unit tests, fixes any errors there, runs the server, uses curl to test the change live.
It isn't perfect, but it takes about ten minutes to do the above. I trust my experience to review the work, and when AI produces bad work, I just move on, maybe keeping some of it.
I also see a lot of AI slop, including from equally seasoned engineers working in the same stack, so my unproven hunch is that there's a lot of learning curve to using it effectively. But I can't argue that it makes me a lot faster, even after accounting for the time I spend reviewing and fixing mistakes.
The number of Android app releases is likely going down because Google is increasing minimum requirements for publishers. They require them to be updated to newer versions of the OS, etc. It seems...
The number of Android app releases is likely going down because Google is increasing minimum requirements for publishers. They require them to be updated to newer versions of the OS, etc. It seems they don't want a huge number of barely maintained, similar apps in the app store anymore?
I don't know about the other app stores, but they might be limited by gatekeeping, too? Apple manually reviews updates and they probably have limited capacity.
These days Play Store requirements are becoming more onerous than those of the App Store, which I don’t think was on anybody’s bingo card. You basically have to ship an update incrementing the...
These days Play Store requirements are becoming more onerous than those of the App Store, which I don’t think was on anybody’s bingo card. You basically have to ship an update incrementing the targeted Android SDK at least once a year, which usually involves several breaking changes. It’s a fair chunk of work for someone publishing an app as a hobby.
By contrast you can let your apps go without updates for at least a couple years on the App Store, and AFAIK there’s no minimum SDK requirement, so if you wanted to you could probably get by for a few years with a frozen Xcode setup that you use to publish updates that do nothing but increment the version number.
Another big issue is how EU regulations now require indie devs to doxx themselves by putting their home address on the App/Play Store listing page. You can get around this by creating an LLC with its own separate registered address, but again that’s a lot of overhead for a hobbyist or FOSS volunteer.
I'm reminded of how, to send an email newsletter, anti-spam regulations apparently require you to have a postal address and include it in the email. When setting up a Substack account, there's a...
I'm reminded of how, to send an email newsletter, anti-spam regulations apparently require you to have a postal address and include it in the email. When setting up a Substack account, there's a box for this, but it's prefilled with Substack's mailing address. I guess if they ever get any mail for you, they will let you know?
Perhaps some middleman will make publishing Android apps easier.
I believe the responsible legal party has to list their address. If someone wanted to be the middleman for anonymous developers, they'd likely be assuming legal responsibility, the same as an LLC.
I believe the responsible legal party has to list their address. If someone wanted to be the middleman for anonymous developers, they'd likely be assuming legal responsibility, the same as an LLC.
Regardless, if AI coding assistants really made coding 10x faster, surely it wouldn't be too difficult to meet those requirements? You'd think that if what these AI companies were saying was true,...
Regardless, if AI coding assistants really made coding 10x faster, surely it wouldn't be too difficult to meet those requirements?
You'd think that if what these AI companies were saying was true, maintaining compatibility with new android versions would be far easier now, not harder.
A great blog post that applied Occam's Razor to AI coding claims, and then backed that up with some data.
The premise is simple: If AI programming assistance was really increasing productivity as much as reported, there would be a gargantuan increase in toy projects and a flooding of all app stores with garbage since 2020.
And by all accounts, that hasn't happened. And it adds a tragic cap to the rounds of layoffs occuring because of the bullshit productivity claims.
It's stupid to fire engineers for not using AI coding tools, but I think that's basically post-hoc justification.
In general, the C suite is looking for an excuse to do layoffs. If revenue isn't growing, the only way to increase profit is cutting costs. When a company does layoffs, they can be honest and say they're cutting headcount and likely losing productive capacity, or they can lie about massive productivity growth through AI tooling.
It isn't much different from doing layoffs through RTO policies. The evidence on productivity isn't there, but executives will gladly lie through their teeth to shirk the blame.
Or they can say nothing, because they're not obligated to give a reason. Which is what they tend to do in pretty much every case.
Most of these "X is doing layoffs because of Y" headlines are heavily mined quotes from interviews of people who may not even be at a leadership level where they had any control over it very much after the fact.
I feel like I see a lot of random anthropomorphizing of companies. Price increases and layoffs are a common one. Companies don't need to, nor do they, talk to anyone when they do either. They don't need an excuse.
They sure are obligated to give a reason to their shareholders and they almost always do. The topic frequently comes up in earnings calls. It just so happens that efficiency gains as a reason shows a lot more certainty than layoffs do.
Even moreso in Europe. They can't fire me without cause and must have a good reason to lay me off that needs to be signed off by a work council. In fact I was recently laid off and was first told their plans for the future and the reason why they were "making the organisation leaner" before they said the actual words that I was fired.
I know that my US counterparts were usually told less and escorted out of the door the minute after the meeting, something absolutely not allowed here, but even the US org needed to provide a reason to their shareholders.
I think the shareholders would be plenty happy with the supposed "real" reason
There are two sides to the labor coin. The cost, and the productivity. Shareholders love hearing about the cost part: "we slashed our operating costs by 60% by firing 1000 developers!!!" They don't like to hear about the productivity part: "our new feature releases will now take 5 times as long".
AI lets them pretend that they get to do the first part without the second part.
While that’s very true in a basic sense, I’d wager the hype-like promise of “10x productivity must mean 10x revenue soon!!” would sell better, since it doesn’t imply stagnant revenues, so that is probably what firms prefer for outward communication.
A little while ago this was posted on tildes: No, AI is not making engineers 10x as productive: curing your AI 10x engineer imposter syndrome
I think it complements this article quite nicely by also going over the motivation and perspective of people making claims.
As I have said in there, while I do get use out of LLMs most of my use cases are underwhelming compared to the hype.
That really nails it for me. The hype has really made it impossible to discuss how these types of programs are best used. It’s all focused on these fantastical, future claims of massive social change.
It does seem like normal, non-tech people are starting to get annoyed at having AI shoved in every single product. So hopefully we’ve reached the crest of the wave and it can come crashing down.
AI is in a weird place for me as a developer. It does sometimes make certain aspects of my work faster, in the moment, but usually it also means I miss out on a whole research/discovery process that massively increases my knowledge of a subject, which in turn means other work now becomes slower because I didn't learn as much.
Partly because of that, and because I tend to not trust the AI, I end up using it as a suggestion towards a solution and then look to stackoverflow or the docs to back it up, and make sure i understand it all. That's not really efficient.
Its good for a few things
Helps with best practices in a new language (ex how to do blah general task in Go)
Helps newbies learn quicker by generating specific answers without having to wait for someone to respond (why does python use if name == main? With free followup questions
Builds good templates for things like yml files
Does a pretty good job of reading documentation for you, but when its wrong, boy is it wrong.
I don’t think a developer with over 20 years experience is using it much for these things, the things that its good at.
Idk what these senior guys are using it for, really, cause we all know it doesn’t work great within complex codebases, too much needed context, and Im convinced anyone who says that it can do that is lying.
Then you would be wrong. I have over 25 years of experience and these days I use Claude Code almost full-time.
That is because I am currently working in languages that I don't really know, but that are related to languages/technologies that I do know. Last week I spent two days setting up a (for me) complicated cloud configuration on GCP with Terraform. I have experience with Pulumi, I have experience with configuring those types of environments on GCP and AWS, but I had actually never used Terraform before. I can explain in clear terms what I want from Claude and I know which questions to ask when I read the code it spits out. In addition it turns out Claude is incredibly fast at writing gcloud commands that help debug something that isn't quite right. Really, if I had had to learn all that from the documentation, it would have taken me days extra, if not a week.
The weeks before that I was writing web services in Python. Same thing. I have over two decades of experience in writing internet backends, but somehow I have never done one in Python. I know how to ask for what I want and I know how to test and debug what I get back. Of course, I could have written it all by myself, and it would not have taken me months or weeks. But simply not having to worry about the exact syntax, or figuring out what the standard libraries for X or Y are, is a massive timesaver.
I think plenty of seniors end up on a wide variety of projects. I've seen seniors with too much bluster try to learn a front-end framework in a weekend...
I'm not sure how to respond to your last sentence, but I'll take the challenge. I'm not an AI priponent, but I use it at work and it definitely makes me faster. I have twenty five years as a software engineer.
My codebase is a well organized monorepo of related services and tools. They're joined together with an open API spec, strict typescript types, and a DB schema. I mention those because it think AI does best with strong metadata about the code, which these things are. I also have a lot of unit tests.
If I ask for an update, I usually describe it in terms of a DB schema update and a corresponding API change. AI usually does a good job of solving for me. It updates the spec and schema first, then the rest flows nicely. It makes my changes, fixes typing errors, runs unit tests, fixes any errors there, runs the server, uses curl to test the change live.
It isn't perfect, but it takes about ten minutes to do the above. I trust my experience to review the work, and when AI produces bad work, I just move on, maybe keeping some of it.
I also see a lot of AI slop, including from equally seasoned engineers working in the same stack, so my unproven hunch is that there's a lot of learning curve to using it effectively. But I can't argue that it makes me a lot faster, even after accounting for the time I spend reviewing and fixing mistakes.
The number of Android app releases is likely going down because Google is increasing minimum requirements for publishers. They require them to be updated to newer versions of the OS, etc. It seems they don't want a huge number of barely maintained, similar apps in the app store anymore?
I don't know about the other app stores, but they might be limited by gatekeeping, too? Apple manually reviews updates and they probably have limited capacity.
For the web, perhaps SEO is harder nowadays.
These days Play Store requirements are becoming more onerous than those of the App Store, which I don’t think was on anybody’s bingo card. You basically have to ship an update incrementing the targeted Android SDK at least once a year, which usually involves several breaking changes. It’s a fair chunk of work for someone publishing an app as a hobby.
By contrast you can let your apps go without updates for at least a couple years on the App Store, and AFAIK there’s no minimum SDK requirement, so if you wanted to you could probably get by for a few years with a frozen Xcode setup that you use to publish updates that do nothing but increment the version number.
Another big issue is how EU regulations now require indie devs to doxx themselves by putting their home address on the App/Play Store listing page. You can get around this by creating an LLC with its own separate registered address, but again that’s a lot of overhead for a hobbyist or FOSS volunteer.
I'm reminded of how, to send an email newsletter, anti-spam regulations apparently require you to have a postal address and include it in the email. When setting up a Substack account, there's a box for this, but it's prefilled with Substack's mailing address. I guess if they ever get any mail for you, they will let you know?
Perhaps some middleman will make publishing Android apps easier.
I believe the responsible legal party has to list their address. If someone wanted to be the middleman for anonymous developers, they'd likely be assuming legal responsibility, the same as an LLC.
Regardless, if AI coding assistants really made coding 10x faster, surely it wouldn't be too difficult to meet those requirements?
You'd think that if what these AI companies were saying was true, maintaining compatibility with new android versions would be far easier now, not harder.