If you never write the code, you'll never be able to read the code effectively. Also, you will not pick up the ability to reason algorithmically. This is why I tell students that they must write the code in my classes:
Leaving aside the power you are giving to the AI companies by entirely giving up on coding. I say this as someone who uses AI extensively, if ambivalently.
I see the same thinking in AI-art boosterism: having "good ideas" is the important and "creative" part while actually implementing them is drudgery best delegated to someone (or something) else.
Huh? You presented zero arguments. You didn't explain why the take is "hyperbolic". And you pretentiously posted a link about how the writing is AI assisted without addressing any of the underlying arguments in the writing. As if using AI to write is a bad thing.
Your comment is a waste of time.
ps. Not to mention the fact that you seem to be unaware of the fact that it's theoretically impossible to discern human writing from LLM writing unless there is an explicit signature in the writing.
It is, it's disrespectful of the reader. Like others have said time and again across HN: why should I spend my time reading an article the "author" couldn't even spend their own time writing?
> Not to mention the fact that you seem to be unaware of the fact that it's theoretically impossible to discern human writing from LLM writing unless there is an explicit signature in the writing.
Because with code the reviewer most likely believes you are putting more effort into writing code with an AI than you did handwriting it. If you don’t then your coworkers most likely already hate reviewing your agents crap.
The plot labeled “Incoming bug reports over the same period.” is funny. Where does the red line come from, except arbitrarily setting 20 and 15 bug reports and plotting that? There’s no discernible connection to the actual data.
Even if the numbers made sense it would be pretty damning for OP's company. He's outright admitting that Filestage's engineering team is so dysfunctional that the code they were producing is even worse than slop?
I’m confused how LLM code is so different from open source code, from the perspective of the developer.
Like if someone gave you a magic wand that let you conjure a mature open source library that did most of what you needed for your project, that would be amazing right? Would you refuse to use it because you miss writing your own database or API framework? Or would you just get to build way more ambitious things?
Software is already stitching a bunch of things together in most cases, why not stitch together things made by a machine instead of by an open source developer? What percent of developers primarily work on bare metal algorithms? .0001%? Everything else is stitching together existing pieces at some level of abstraction.
The real difference seems to be the non-determinism, not that the code is produced automatically. And I think a lot of the work right now that people don’t like seems to be people dealing with the non-determinism, not that so much code can now get generated.
I have read accounts of people have similar freak outs about higher level languages, object oriented programming and web services, and all of those just made more demand for software because the potential of software has not nearly been reached. I wonder if the extreme growth of the field has resulted in most people feeling like those huge changes that reduced the amount of code that had to be written per unit of functionality have been around since the dawn of time, and now that another level of code generation has been reached it’s the end of the world (again).
Well cause I don't really care about the code so much as I care that there's someone or some group maintaining it. You've never looked at the last commit time when you want to include some library in your project? I need proof of some buy-in from someone else. That someone is putting their time on the line. Otherwise I know using that package is gonna cost more time than its worth.
I have a hard time believing the author’s team is doing all their coding with AI, with only the $20/month tier of subscription. When I experimented with writing zero code by hand, my token spend shot up to more like $300/mo.
I’m also not ready yet to conclude that that option, as opposed to a more modest version of AI assisted coding, will remain open to me indefinitely. We now have financials confirming that Anthropoic, too, is running a -200% profit margin, which leaves serious cause for concern about whether either GPT or Claude models will still be around in another year or two. And the judge is still out on whether local models can perform well enough to actually do all my coding for me.
I question the epistemology of people who did not realize this by around late-2024. (Sonnet 3.5 was out and could do some things okay.)
It's defensible to say "October 2026 is not yet the time when we should be 100% deferring code writing to AI (let alone not even looking at the code any longer)". But the writing is very obviously on the wall. It's likely not going to take longer than 10 years.
So many of these pro-AI articles lean on references to "people used to be scared of calculators!" and the like, but it's a nonsense comparison. Calculators weren't predicated on monetizing prior work at the cost of devaluing future work. Functioning calculators are at least as accurate as a human, generally more so, and they do not produce incorrect outputs when given correct inputs no matter what. And calculators don't lie to you about what they've done, even if the result will make you sad. The comparison doesn't hold up.
(Also, I think some of the concerns about calculators at those times actually were valid, even though I wouldn't argue that society should not have adopted their use.)
https://htmx.org/essays/yes-and/
Leaving aside the power you are giving to the AI companies by entirely giving up on coding. I say this as someone who uses AI extensively, if ambivalently.
Don’t waste your time.
I hope these hyperbolic takes can be over, forever.
I wonder if the author also agrees with DHH’s racist opinions.
Your comment is a waste of time.
ps. Not to mention the fact that you seem to be unaware of the fact that it's theoretically impossible to discern human writing from LLM writing unless there is an explicit signature in the writing.
Which seems like a pretty good argument to me, given the amount of slip out there and how increasingly performative these pieces are.
It is, it's disrespectful of the reader. Like others have said time and again across HN: why should I spend my time reading an article the "author" couldn't even spend their own time writing?
> Not to mention the fact that you seem to be unaware of the fact that it's theoretically impossible to discern human writing from LLM writing unless there is an explicit signature in the writing.
Huh? No it's not lol
The same reasons you might spend your cycles running code the "author" couldn't even spend their own time writing.
With the right high-level abstractions, I can write code as fast as trying to convert that thought into English (for some tasks).
How fast can you scaffold new ideas and delete them when they don't work?
Like if someone gave you a magic wand that let you conjure a mature open source library that did most of what you needed for your project, that would be amazing right? Would you refuse to use it because you miss writing your own database or API framework? Or would you just get to build way more ambitious things?
Software is already stitching a bunch of things together in most cases, why not stitch together things made by a machine instead of by an open source developer? What percent of developers primarily work on bare metal algorithms? .0001%? Everything else is stitching together existing pieces at some level of abstraction.
The real difference seems to be the non-determinism, not that the code is produced automatically. And I think a lot of the work right now that people don’t like seems to be people dealing with the non-determinism, not that so much code can now get generated.
I have read accounts of people have similar freak outs about higher level languages, object oriented programming and web services, and all of those just made more demand for software because the potential of software has not nearly been reached. I wonder if the extreme growth of the field has resulted in most people feeling like those huge changes that reduced the amount of code that had to be written per unit of functionality have been around since the dawn of time, and now that another level of code generation has been reached it’s the end of the world (again).
I’m also not ready yet to conclude that that option, as opposed to a more modest version of AI assisted coding, will remain open to me indefinitely. We now have financials confirming that Anthropoic, too, is running a -200% profit margin, which leaves serious cause for concern about whether either GPT or Claude models will still be around in another year or two. And the judge is still out on whether local models can perform well enough to actually do all my coding for me.
It's defensible to say "October 2026 is not yet the time when we should be 100% deferring code writing to AI (let alone not even looking at the code any longer)". But the writing is very obviously on the wall. It's likely not going to take longer than 10 years.
Then we saw the capabilities really take off, and it was obvious how things would go.
(Also, I think some of the concerns about calculators at those times actually were valid, even though I wouldn't argue that society should not have adopted their use.)
and then, at some point, that will cease to be able to happen. they'll be good enough.
i predict that happens before March 2027.