This resonates. I've shipped 7 side projects in the past year using AI heavily. The code writing part got 10x faster. But I've noticed something counterintuitive: the total time from idea to shipped product barely decreased.
Why? Because the bottleneck was never typing code. It was always understanding the problem, making architectural decisions, debugging edge cases, and most importantly - knowing what NOT to build.
AI made me faster at producing code, but it also made me produce MORE code, which means more surface area for bugs, more maintenance burden, more complexity to reason about. The discipline of "write less code" is harder now because writing code costs almost nothing.
The engineers who thrive will be the ones who can resist the temptation to over-engineer when the marginal cost of adding complexity drops to near zero.
> Why? Because the bottleneck was never typing code. It was always understanding the problem, making architectural decisions, debugging edge cases, and most importantly - knowing what NOT to build.
For me, this is a bit different. Writing code has always been the bottleneck. I get most of my joy out of solving edge cases and finding optimizations. My favorite projects are when I’m given an existing codebase with the task, “When mars and venus are opposite eachother, the code gets this weird bug that we can’t reproduce.”
When a project requires me to start from scratch, it takes me a lot longer than most other people. Once I’ve thought of the architecture, I get bored with writing the implementation.
AI has made this _a lot_ easier for me.
I think the engineers who thrive wi be the ones know when to use what tool. This has been the case before AI, AI is just another tool allowing more people to thrive.
Same here. I do well in existing codebases because I can follow patterns and adapt to existing limitations but starting a new project is always so daunting to me. Writing a spec and iterating on it is so much more natural than writing code in a new project for me.
Fascinating- I've always loved the big picture, architecture, and I've also loved stable software but have one hell of a time fixing bugs. AI helped me a ton with that.
Well if you're ever in need for a complementary mind in side projects- huh, how does one connect over HackerNews?
I have my own side project that I vibe coded. I probably did what would take one team 6 montns and produced it myself in one month.
I'm not afraid of breaking stuff because it is only a small set of users. However for my own code for my professional job no way I would go that fast because I would impact millions of users.
It is insane that companies think they can replace teams wholesale while maintaining quality.
> It is insane that companies think they can replace teams wholesale while maintaining quality.
The assumption is that AI will continue to improve. If we get another one or two quality jumps over the next 1-3 years, which is not totally unreasonable, AI quality might be good enough.
Yeah I vibe coded an addition game for my 4 year old that lets him do addition problems where the answer is always 10 or less. It’s very “juicy”. There’s a lot of screen shake and spinning and flashy rainbow insanity going on. If I had done all that stuff myself it would have take a week because I would have been picky about each little animation. The thing that saved me the most time was just being ok with the good enough animations the ai spit out.
It’s amazing for him and it works on his iPad.
However when I tried it on my iPhone it was a broken mess. Completely unusable (not because of screen size differences).
I tried getting Claude to fix it but it couldn’t do it without changing too much of the look and feel, so I dug into the code and it was thousands of lines of absolute madness. I know from using this at work that there are things I could have done. Write tests to lock in things I like etc…
But so much of the speed up was about not caring about the specifics that once I started caring about making an actual product, I was not much faster maybe not any faster at all. The bottleneck in writing a game was never in banging out code.
> I dug into the code and it was thousands of lines of absolute madness
Ask the AI to assess the code itself and to propose ways to gradually refactor it for better cleanliness. It can be good at that stuff, but you need to make it an explicit goal.
Yeah I tried that but without tests it couldn’t keep the look and feel the same. And spending time thinking deeply enough about it to understand and specify what exactly I don’t want it to change just goes back to my point that coding isn’t the hard part.
> I probably did what would take one team 6 montns and produced it myself in one month.
I find it… Amusing? That’s not quite the word. That programmers—a group notoriously for making wrong estimates of how long something will take to build—continuously and confidently spew a version of this.
And it’s not even estimating how long we ourselves would take to build something, now we’re onto estimating what an undetermined team of completely made up strangers could do. It’s bonkers. It has no basis in reality.
It’s not an estimate, the point was that AI produced code multiples faster than the prompter, and the prompter is in a pretty good position to make that claim. I can confirm and make the same claim, so I believe that it’s true that for some tasks, Claude makes me 10x faster than on my own without AI, where 10x absolutely is a completely made up number that’s still true in spirit.
Yes, it is. “It would take a team 6 months” is an estimate, and I don’t see how you can argue it’s not. Even if it just said it would take them longer, that would still be an estimate.
> Claude makes me 10x faster than on my own without AI
Also an estimate.
> where 10x absolutely is a completely made up number
And by your own admission, an estimate taken from the ass that you thus cannot be certain is true. Made up perception does not equal reality.
Yes you do, you already made the argument when you pointed out the “team” size and makeup was completely unspecified, therefore the number is not an estimate, it’s just a number.
When you call it an “estimate” you are adding additional unsupported specificity to something that was explicitly stated as being hand-wavy to make an obviously rhetorical point. Are you saying you can’t understand the point being made?
My 10x is based on my experience doing projects with Claude. I also said “some” tasks, not all tasks, and I didn’t specify which tasks, and I clarified that my number is made up, which is why my number is also not an estimate of anything. There are some tasks that Claude can do 10x faster than me, and there are some tasks that it can do 100x faster than me, and there are some tasks I can do faster than Claude... for now... More importantly for me personally, Claude makes starting projects and using tech I don’t already know easier; it’s lower effort, regardless of speed.
The paper is interesting and a valid data point, but I don’t think it proves your point. I’ll respond with a few thoughts.
First, the dev’s self estimate of AI productivity speedup was +20%, even though their measured productivity was -20%. This may relate to the effort and not the speed, and it’s important to note that this is a gray zone the paper didn’t explore, and something that can be true on both sides. I can be “faster” at developing and still take the same or longer wall clock time. Measuring the time doesn’t capture how the time was spent, nor the qualities of that time.
Second, this study was done a year ago. That’s an eternity in AI land, and everyone noticed Claude and other models getting substantially better at code writing last fall, plus workflows and tooling are improving even faster than that. There’s every reason to believe the outcome of the exact same study might be different this year than it was last year.
Third, this study is explicitly biased toward large projects, and large projects are, even today, more difficult to find the productivity boosts with. I find Claude absolutely amazing at starting new projects, and absolutely terrible at working in large code bases that don’t fit in context. When I say Claude makes me 10x faster at some projects, I’m referring to something like setting up a new CRUD app when I don’t know much about setting up a database and web server backend, or writing a graphics app in Vulkan when I’ve only used OpenGL. Doing stuff like that, having Claude help me with tech stacks I don’t know, absolutely is many multiples faster than doing it on my own, and the paper link you’ve shared doesn’t address that use of AI at all.
Note specifically the paper says they are not demonstrating or claiming that “AI systems do not currently speed up many or most software developers”, and they have not demonstrated or claimed that “AI systems in the near future will not speed up developers in our exact setting”. It might be a mistake on your part to try to use this as some kind of evidence that AI isn’t speeding devs up.
A missing link right now is automated high-quality code reviews. I would love an adversarial code review agent that has a persona oriented around all incoming code being slop, that leverages a wealth of knowledge (both manually written by the team and/or aggregated from previous/historical code reviews). And that agent should pull no punches when reviewing code.
This would augment actual engineer code reviews and help deal with volume.
The issue is that before AI, 1% of the population was capable of creating 1 side project per year. After AI, 10% of the population is capable of creating 10 side projects per year. The competition grew by 100x. The pessimist in me thinks that the window of opportunity to create something successful is shrinking.
> The pessimist in me thinks that the window of opportunity to create something successful is shrinking.
Dunno man. Ideas alone aren't worth anything [0] and execution is everything [1], but good ideas and great execution will never go out of style regardless of how much competition is out there. I'm of the opinion that even if 10% of the population is now capable of creating a side project, there's still the same relatively-fixed amount of people capable of making a good side project, and even fewer who will see it through to a real product. Nothing has really changed in the aggregate. It's like architecture, there are always improvements in materials, tools and processes, and Claude and Codex can provide more laborers for almost free, but most people are still gonna be building uninspired McMansions instead of the Guggenheim.
Disagree. Ideas were a necessary component of the one project I had success with. BTW, the line between ideas and execution is blurred. Is coming up with innovative UI and features ideas or execution?
Why do you look at it that way? Why does anyone beside you have to care about what you do?
Just build something for yourself. You will always have things you'd like to build for yourself. You will be in competition with yourself only and your target audience will be yourself.
Market forces do not apply to side-projects, because that's what people do for fun.
Just because there are chess computers, doesn't mean that no one plays chess anymore at home.
Isn't it obvious? The reward that a personal project can generate for you is limited. It's not remotely close to what a successful project would give you - money, fulfillment, social capital, feeling good about yourself, etc.
It was wrong to write software you hoped others would use? The entire open source ecosystem works on this idea otherwise there would be no point in sharing and we can move to closed software.
Yeah but we've told ourselves that writing software was some kind of higher mathematics, where in reality it was mostly just plumbing that, surprise, a computer can do too.
> It was wrong to write software you hoped others would use?
Yes.
> The entire open source ecosystem works on this idea otherwise there would be no point in sharing and we can move to closed software.
No.
The _actual_ open source system consisted of hackers scratching their own itch and sharing the artifacts, because (it was assumed that) sharing is free. So if the work is already done and solved their problem, why not also share it as gift.
This remains unchanged.
The driving force of FOSS is not "how can I fix someone else's problem". It never has been.
Well.. maybe on HN it was different, but that's not "the open source ecosystem". And, yes, maybe some corps have gaslit naive people into believing that they must donate their lives to said corps.
> The _actual_ open source system consisted of hackers scratching their own itch and sharing the artifacts, because (it was assumed that) sharing is free. So if the work is already done and solved their problem, why not also share it as gift.
If you have the time tona scratch your own itch and gift the results, it implies you have a source of income that gives you the time/lifestyle to do such a thing. You might be a tenured academic, or live in a society with a strong safety net. Or you might be able to do your day job in 1/2 the allotted time.
The problem is that a those scenarios are eroding precipitously, leaving more to seek compensation for their work output, whether it is closed or open source.
Not only that, I have a feeling a lot of people are gonna be disappointed now they can implement their side projects in a week instead of 6 months. Finally - the thing is there, ready. And the likely outcome is
a) Almost no one but you cares and
b) Now that this has become trivial, there's no much joy in it. The struggle we had before A.I was the real joy; prompting agents for a few days and getting what you want isn't that joyful.
Ironically I had a very smart and otherwise reasonable math professor who, shortly after Kasparov lost to Deep Blue, said in class that chess was no longer interesting.
Maybe, but LLMs solve but one issue (maybe two). Take me, for example. I am highly proficient regarding software development in most aspects. Except for that tiny problem: I wouldn't even know what to build. And at least for me, LLMs could not help with that.
The whole side project or even private project thing doesn't just hinge on being able to produce software. There's a lot more.
It's like the business of selling electric drills. People don't really want drills they want holes. But holes are difficult to sell so the selling the drills is a proxy for that.
In software it's the same thing. People don't really want software they want data and data transformation. But traditionally the proxy for that has been selling the software (either as a desktop app or then later as sole kind of service).
You could argue that in either case the proxy is not what people want but yet because of the difficulty of selling the "actual" thing the proxy market has flourished.
We're now inventing a new tool that will completely disrupt that market and any software business that is predicated on the complexity required to create the software to transform the data is going to get severely disrupted. Software itself will be worthless.
The value of computers since its inception was that it's capable of transforming data very, very fast and autonomously. But someone has to input that data from the real world or capture it using some device, and someone has to write the rules.
What happened is that we created a whole world of information and the rules has become very complex. Now we have multiple layers stacked vertically and multiple domains spread horizontally. At one time, ASCII was enough, now we have to deal with Unicode.
Software becoming worthless will mean that everyone has learned the rules of the systems we created and capable of creating systems with good enough quality. I'm not seeing that happens anytime soon.
Software is just means to an end. Data and data transformation is what people want. Software has sellable dollar value only because creating the software to do the data transformation has had real associated cost. I.e anyone who wanted a particular data transformation had to pay to get the software that does it.
When you drive down that cost you drive down the potential value of the software products. Remember that what is a cost to one party is revenue to the other party. Without revenue there cannot be profit and without revenue software has no dollar value.
If anyone can create "photoshop" with minimal cost and there are thousands of said "photoshop" apps what will be the retail sell value of those apps. Close to zero.
This same lifecycle already happened with games. Driving down the cost of producing games resulted in a proliferation of games that are mostly worthless that you can't even give away.
> Software is just means to an end. Data and data transformation is what people want. Software has sellable dollar value only because creating the software to do the data transformation has had real associated cost. I.e anyone who wanted a particular data transformation had to pay to get the software that does it.
I do agree with you on that point.
> If anyone can create "photoshop" with minimal cost and there are thousands of said "photoshop" apps what will be the retail sell value of those apps. Close to zero.
This is the point that I cannot agree with. Not anyone can create photoshop because of the amount of knowledge you need about the data and transformations that needs to be applied to get a specific result. And then make a coherent system around it. You can create isolated function just fine, just like a lot of people knows how to build a shed with planks and nails. But even when given all the materials and tools, only a few can build a skyscraper or a mansion.
That knowledge of how to create a coherent systems that does something well is the real cost of software. Producing code isn't it.
You're right and I agree with you to an extent. Also as of now the tools aren't quite intelligent enough for one to produce software of that complexity without having someone competent at the helm.
That being said what already exists was already enough to shutter the stock prices of many software companies precisely because the fear is that their clients will just re-create the software themselves instead of buying it from someone else.
I guess we'll see how this will pan out in the next few years.
Yes it become much easier to fail fast and iterate, but also a lot of these fail fast projects are trivial for anyone to implement themselves. Differentiating your project is going to be tougher too.
A lot of the moats are gone, but quality (and security) is in a nose dive. AI built project might be the Ikea furniture. Good for the masses, but there's still a market (much smaller) for well crafted applications and services. It's hard to say what it'll look like in a couples years though. Maybe even the crafting is eventually gone. /shrug
But the total market size (in number of products) also multiplied. For instance, as a relatively tiny example, I create a nutrition tracker. There's hundreds already out there, but they never met my specific desire for one. So I created one with Claude (took maybe 2 hours total over a few days) that completely matches my desire, plus I can tweak it as want for my needs.
No one else will want this specific piece of software. But I love it.
Sure, there will be 100x the competition, but there will be also 100x the software needs. Now, if you want to get crazy rich building software, that does get tougher, but that's a good thing, I think.
If the competition just grew by 100x, where's all the great, high-quality, AI-vibe coded side products? Something just isn't adding up here. Could it be that vibe coding on its own just isn't all that useful, and most of those 10% are wasting their time?
The counterpoint is that it's only 2 months since AI got really useful and it will presumably continue to improve. It takes a while until it spreads through the society.
Are most side projects in competition? I wouldn't think so.
Even if they were I disagree that 10x more ideas being produced means 10x more products in competition. You could leverage AI to execute but still have terrible ideas, leadership, product stewardship etc.
I think some clever people with a real and valuable insight will finally be able to turn that insight into a product. I also think the other 9 products will be get rich quick attempts by people with nothing to offer.
I think we need to change our perspective of what success is. I believe there will be a ton of small companies popping up instead of a few big ones that eats everyone's lunch. Like Google, Microsoft and others giants have done until now.
I can relate. Sincerely debating whether I quit my well-paying and comfortable corporate job and just go full-time entrepreneur before the opportunities disappear.
The game is all about content now. Forget software. Games, movies, books, music, etc. Things that people will always want regardless of how much there already is. Look at the success of AI slop authors and YouTube channels. That's our future.
> I've shipped 7 side projects in the past year using AI heavily. But I've noticed something counterintuitive: the total time from idea to shipped product barely decreased.
> Why? Because the bottleneck was never typing code.
Were you also shipping side projects every 2 months before AI?
If not, this comment just reads like cognitive dissonance. Your core claim is that AI has enabled you to ship 7 projects in 12 months, which presumably was not something you did pre-AI, right? So the AI is helping ship projects faster?
I agree that AI is not a panacea and a skilled developer is required. I also agree that it can become a trap to produce a lot of bad code if you’re not paying attention (something a lot of companies are going to discover in 2026 IMO)
But I don’t know how you can claim AI isn’t helping you ship faster right after telling us AI is helping you ship faster.
I could see that being the case in a company where you’re waiting on stakeholders and other people, but the parent commenter was talking about their personal side projects.
The new bottleneck is code ownership. You have to understand what it does and how it works to maintain it long term. You can LLM into a maintainability disaster but you can’t LLM out of it. Biting off more than you can chew is more dangerous than ever.
We need to have more metrics for this. Like I hear people making this claim on HN all the time as if they know absolutely for sure but I doubt it's this simple.
I can guarantee you this... the story is not absolute. Depending on who you are and what you need to work on dev time could be slower, same or faster for you. BUT what we don't know is the proportion. Is it faster for 60% of people? 70%, 80%?
This is something we don't know for sure yet. But i suspect your instinct is completely wrong and that 90% of people are overall faster... much faster. I do agree that it produces more bugs and more maintenance hurdles but it is that much faster.
The thing is LLMs can bug squash too. AND they are often much faster at it then humans. My agentic set up just reads the incoming slack messages on the issue, makes a ticket, fixes the code and creates a PR in one shot.
I totally agree, except the more we get used to working with the tools the better and faster things will get. I would argue the field has been evolving fast in the past 3 years, but now it's showing signs of slowing down. And I think this is good, as it will allow people to catch up, and refine the approach to adapt better to the new paradigm of coding.
Fortunately, AI can also be used to reduce complexity. The case I noticed most often is to use the slightly more ugly API, or duplicate some generic code, but avoid pulling in a dependency. Examples are avoiding UI frameworks and directly accessing the DOM in simple web projects, using the CLI arg parser from the stdlib or adding simple helper functions rather than pulling in left-pad like dependencies.
Since managing dependencies is one of the major maintenance burdens in some of my projects (updating them, keeping their APIs in mind, complexity due to overgeneralization), this can help quite a lot.
Well how many side projects did you ship last year? I’ve written small programs in the last few months over a weekend that would have taken me a month to do a couple years ago, and they’re better. Not in terms of code quality, but in terms of features I wanted and knew how to implement but couldn’t be bothered, Opus can do in one minute and even if it’s not the optimal implementation it’s completely functional, fine, and costs me almost nothing.
When the goal is to ship (the result) I'll happily leverage LLM's to try an idea or 3 out. However, it wouldn't be fair to claim that my side projects have exactly one goal. That's why I choose to use AI generated code when I deal with stuff that I already know how to do, done a lot of times, and the only thing that I gain from using AI is time typing it out.
Anything else? I'll struggle and grow as a developer, thanks. And before anyone says "but there are architecture decisions etc. so you still grow"... those existed anyways. If I have to practice, I'll practice micro AND macro skills.
This tracks with the way a lot of heavily vibecoded projects have issues with beeing feature heavy, while those features often don't fully work and most importantly don't fit together cohesively. In other words, the quality is low.
> Because the bottleneck was never typing code. It was always understanding the problem, making architectural decisions, debugging edge cases, and most importantly - knowing what NOT to build.
The AI can help you in these tasks too, but you need to ask for the help in terms that it can help you with, and not expect it to be genuinely intelligent or to have a crystal ball. As a bonus, once you've gotten these things into the agentic context, the code itself becomes better too.
> AI made me faster at producing code, but it also made me produce MORE code, which means more surface area for bugs, more maintenance burden, more complexity to reason about
I think from time to time, it's better to ask the AI whether the codebase could be cleaned and simplified. Much better if you use different AI than what you use to make the project.
They've always said you spend a lot more time reading code than writing it. If suddenly you're writing a lot more code, you're going to spend a ton more time reading it.
When I got to the part where it said that developers chose software engineering as a job because they like to code not because they want to review or "manage" code I really felt that. But while I enjoy coding & building as solo developer on my projects I can't really say I've ever enjoyed it as a job. Or are you not supposed to like your job? Is that how the world works?
You are putting sentences together just like an LLM would - quite fitting for an AI generated article. You might want to get it checked out, these days you never know if you are a real person or not.
As you mentioned, scope definition and constraints play a major role but ensuring that you don't just go for the first slop result but refine it pays off. It helps to have a very clear mental model of feature constraints that doesn't fall prey to scope creep.
There's also a reward for not over thinking it and letting AI bring the solutions to you. The outcomes are better when it's a question, answer, and execution session.
I get benefits with AI both on the writing the code part and the understanding the problem part. If AI disappeared tomorrow I’d probably still enter “plan mode” in my head. I like having the discussion with the AI about requirements and edge cases and all that, while it updates the plan and documents architectural decisions in CLAUDE.md. I love that I can add extra polish, such as color to terminal output, or other random features that would have not made the cut before. Instead toiling on a random one off script to fix a problem I can have a whole CLI build that is a joy to use. Explaining complex architecture is easy now because instead of a boring EDD I can slop out animations that demonstrate data moving and transforming through a system.
I mean if you've built 7 side projects (and we assume it's the same phase since total time from idea to shipped product barely decreased), how are these things still a bottleneck to you? I'm assuming you're building in a domain/language you're comfortable with by now (unless you're crazy and try something fundamentally different on each of those shipped products).
Why will the 8th project still have those things as the bottleneck given your experience?
Also if you're not seeing any real gains in productivity, why are you using AI for your side projects and wasting tokens/money?
I use AI for side projects because Google gives me a stupid large number of tokens that refresh every 6-24 hours on my existing $10/mo Google One plan. I see it as my civic duty to help increase their costs by producing slop that I generally throw away anyways because it doesn't actually work after it gets generated.
At work, I was told to use AI but it doesn't actually work for anything that I couldn't have handed off to a brand new undergraduate intern. So I use it for things that I don't care about then go spend twice as long rewriting what it output because it made the task longer by being wrong.
> The engineers who thrive will be the ones who can resist the temptation to over-engineer when the marginal cost of adding complexity drops to near zero.
One area --and many may not like that fact-- where it can help greatly is that the cost of adding tests also drops to near zero and that doesn't work against us (because tests are typically way more localized and aren't the maintenance burden production code is). And a some us were lazy and didn't like to write too many tests. Or take generative testing / fuzzy testing: writing the proper generators or fuzzers wasn't always that trivial. Now it could become much easier.
So we may be able to use the AI slop to help us have more correct code. Same for debugging edge cases: models can totally help (I've had case as simple as a cryptic error message which I didn't recognize: passed it + the code to a LLM and it could tell me what the error was).
But yup it's a given that, as you put it, when the marginal cost of adding complexity drops to near zero, we're opening a whole new can of worms.
TFA is AI slop but fundamentally it may not be incorrect: the gigantic amount of generated sloppy code needs to be kept in check and that's where engineering is going to kick in.
Its worth mentioning that this essay has some signs of being either partially AI generated or heavily edited through an LLM. Some of the signs are there (It's not X, it's Y), With the blog having gone from nearly zero activity between 2015 and 2025 to have it explode in posts and text output since then also raises an eyebrow.
The comments as well. I won't give away the tells but HN is less and less pleasant to read. Now is the time to cherish your pockets of small scale high quality forums that's not flooded by this stuff yet.
My current policy on this is that if text expresses opinions or has "I" pronouns attached to it then it's written by me. I don't let LLMs speak for me in this way.
I'll let an LLM update code documentation or even write a README for my project but I'll edit that to ensure it doesn't express opinions or say things like "This is designed to help make code easier to maintain" - because that's an expression of a rationale that the LLM just made up.
I think it is very fair to say that in the same way that LLM's have given english majors access to programming, LLMs have also given engineers access to clear communication.
I'm not shy to admit that LLMs even from 2 years ago could communicate ideas much better than me, especially for a general audience.
It’s not “clear communication” though. The prose that comes out of LLMs is awful - long, vapid paragraphs with distracting tropes. You can ask them to be concise but then they file down all the wrong bits of the sentence and lose meaning. There’s a reason people bother clocking it and complaining about it, it’s *bad*
It’s like everything else that AI can do - looks fine at a glance, or to the inexperienced, but collapses under scrutiny. (By your own admission you’re not a great communicator… how can you tell then?)
It's often warping the message or "snapping it to grid", taking off the edge, the unique insight. A lack of clear communication is much more a symptom of unclarity about the intended message, audience, prioritization etc. I don't doubt that you internally have a clear idea but sharing it requires thinking about the intended audience and the diff of their current state of knowledge and doubt and where you want to move their thinking. This is a much bigger part than knowing eloquent vocab and grammar tricks.
It doesn't come naturally to the more introverted type of person who cares about the object level problem and not whatever anyone else may know or doubt, I'll admit this. But slapping LLMs on it is not a great solution.
> LLMs have also given engineers access to clear communication.
A lot of the time, the inability to express an idea clearly hints at some problem with the underlying idea, or in one's conceptualisation of that idea. Writing is a fantastic way to grapple with those issues, and iron out better and clearer iterations of ideas (or one's understanding thereof).
An LLM, on the other hand, will happily spit out a coherent piece of writing defending any nonsense idea you throw at it. Nothing is learnt, nothing is gained from such "writing" (for either the author or the audience).
Recently read a tweet suggesting to ask an llm to defend a position you know to be false. It's quite eye opening. I mean, it shouldn't be, if you did debate club etc. Or know how lawyers and politicians work. But it's quite revealing how it can piece together a good defense, selectively quoting real facts, embuing them with undue weight etc to make the thesis stand quite well.
As someone who has written a few deeply personal articles with LLM assistance, I see the signs and I'm almost certain this was generated off a few bullet points. The repetition and cadence strongly resembles the LLM output. Its the kind of fluff that I remove from a piece, because it lacks humanity and offers little substance.
Why is this sentiment expressed so often ("It was written/edited by AI"?
It seems to bother people, perhaps since it may have been low-effort.
Doesn't it not matter as long as the content is good? Otherwise, it seems to be no different than a standard low-quality post.
When I want to read Ai writing (which is not never), I chat with it myself and I prompt it better and get more interesting stuff than these generic insight blogspam.
This was my thought after getting through a few paragraphs as well. At first, I was thinking, this is interesting, maybe worth sharing with colleagues. But then it became too obvious it was AI written or "assisted". Can't take that seriously.
Even the linkedin profile has a studio-ghibli-style avatar. People are going to assume that he is just an "analog interface" to an LLM. Which is sad, because he might be a good programmer. In fact, I tend to see a lot of english-as-second-language people embrace LLMs as a kind of "equalizer", not realizing that in 2026 it is the opposite (not saying that it's right either way, just pointing out that it is becoming a kind of anti-marketing, like showing up to a conference without any clothing, and getting banned from the conference permanently).
We should probably normalize publishing things in our native languages, and expecting the audience to run it through a translator. (I have been toying with the idea of writing everything in Esperanto (not my native language, but a favorite) and just posting links to auto-translated English versions where the translation is good enough).
EDIT: as someone with friends and family from Eastern Europe, I can tell you that the prevailing attitude is: "everything is bullshit anyway" (which, to be fair, has a lot of truth to it), and so it is no surprise that people would enthusiastically embrace a pocket-sized bullshit factory, hook it up to a fire-hose, and start spraying. We saw it with spam, and we see it now with slop. It won't stop unless the system stops rewarding it.
LLMs write this way because people write this way. Maybe not everyone, but enough for it to train the models to do it. Much of my writing reads like an LLM wrote it, but that doesn't make me an LLM.
I feel like it's such a lack of self respect and respect for others when people write using AI on personal blogs.
Reading AI code is very pleasant. It's well annotated and consistent - how I like to read code (although not how I write code LOL). Reading language/opinions is not meant to be this way. It becomes repetitive, boring, and feels super derivative. Why would you turn the main way we communicate with each other into a soulless, tedious, chore?
I think with coding it's because I care* about what the robot is doing. But, with communication, I care about what the person is thinking in their mind, not through the interpretation of the robot. Even if the person's mind isn't as strong. At least then I can size the person up - which is the other reason understanding each other is important and ruined when you put a robot in between.
It's also because we (generally) consider a blog to be human communication and we consider math and programs to be something else.
If you're talking to someone on the phone and halfway through they identify themselves as a bot, surprising you, there's a profound sense of something like betrayal. A moment ago you were having a human connection, and suddenly that vaporized. You were misled and were just talking to an unfeeling robot.
And heartfelt writing is similar. We imagine the human at the other side of the screen and we relate. And when we discover it was a bot, no matter how accurate the sentiment, that relationship vanishes.
But with math and software, it's already sterile from a human connection perspective. It's there for a different purpose. Yes, it can be beautiful, but when we read it we don't tend to build a human connection with the coder.
An interesting exception is comments. When we read the fast inverse square root code and see the "what the fuck..." comment, we instantly relate to the person writing the software. If we later learned that comment was generated by an LLM, we'd lose that connection, again.
Totally agree. I'll extend this to email and slacks, too. I cannot stand getting AI written slop from fellow co-workers because they couldn't write the message themselves. Do not even bother to engage with me if you need to put your thoughts through an AI first. It won't go well. People gotta work on themselves a lot more and I think they're using AI to do the opposite.
It is almost 90% generated using AI text. So many paragraphs to say basically nothing at all.
Like look at this paragraph:
> Junior engineers have traditionally learned by doing the simpler, more task-oriented work. Fixing small bugs. Writing straightforward features. Implementing well-defined tickets. This hands-on work built the foundational understanding that eventually allowed them to take on more complex challenges.
The first sentence was enough to convey everything you needed to know, but it kept on adding words in that AI cadence. The entire post is filled with this style of writing, which, even if it is not AI, is extremely annoying to read.
My point is that there's nothing to be written there "instead", it just is not needed text that is added to make the text longer, typical of AI writing that parrots the same points over and over to make up for word count.
Here's another example from the blog:
> Here is something that gets lost in all the excitement about AI productivity: most software engineers became engineers because they love writing code.
> Not managing code. Not reviewing code. Not supervising systems that produce code. Writing it. The act of thinking through a problem, designing a solution, and expressing it precisely in a language that makes a machine do exactly what you intended. That is what drew most of us to this profession. It is a creative act, a form of craftsmanship, and for many engineers, the most satisfying part of their day.
can just be:
> Most software engineers became engineers because they love writing code. It is a creative act, a form of craftsmanship, and for many engineers, the most satisfying part of their day.
Clarity is something that is taught in every writing class but AI generated text always seems to have this weird cadance as follows: The sound is loud. Not a whimper, not a roar, a simple sound that is very loud. And that's why... blah blah blah.
You have to care about your readers if you're writing something seriously. Throwing just a bunch of text that all mean the same thing in your writing is one of the bigger sins you can do, and that's why most people hate reading AI writing.
The part you'd like to remove ("Not managing code...") may be not required to convey the objective meaning of the sentence, but humans have emotions, too. I could have written stuff like that. To build up a bigger emotional picture.
> The act of thinking through a problem, designing a solution, and expressing it precisely in a language that makes a machine do exactly what you intended.
This sentence may not be relevant for whatever you experience to be the relevant message of the text. But it still says something the remaining paragraph does not. And also something I can relate to.
Also, as LLMs are statistical models, one has to assume that they write like this because their training data tells them to. Because humans write like this. Not when they do professional writing maybe, but when they just ramble. Not all blogs are written by professionals. I'd say most aren't. LLM training data consists mostly of humans rambling.
I also sometimes write long comments on the internet. And while I have no example to check, I feel like I do write such sentences, expanding on details to express more emotional context. Because I'm not a robot and I like writing a lot. I think it's a perfectly human thing to do. I find it sad that "writing more than absolutely needed" is now regarded as a sign of AI writing.
One of the good book about writing I read was William Zinsser's "On Writing Well". Striving for simplicity and avoiding clutter was the two first principles described in the book. AI writing feels more like ramblings than communication.
I'm a French speaker and florid and elaborate writing is something I've grown up with. It can be difficult if you don't know the word or are not used to the style, but it's not boring. AI writing is just repetitive.
When I've used AI for proofreading the suggestions it makes to me is to cut a lot and shorten it. It also gives me examples, never with my voice or style though.
5 sentence paragraph. First sentence is parataxis claim. Followed by 3 examples in sentence fragments, missing verbs, that familiar cadence. Then the final sentence, in this case also missing a verb.
It's funny how seemingly easy it is to tell articles like this have that AI generated whiff to them. The first bit that raised my suspicion was the "The Identity Crisis Nobody Talks About" headline. This "The x nobody talks about" feels like such a GenAI thing.
I saw someone point out something like: ai makes every sentence count. There’s no building or allowing a point to breathe. Every sentence is an axiom to get the meaning across, and its so grating
Maybe that's why the writing feels so terrible. The AI is attempting to maximize every sentence while simultaneously expanding on just a few actually meaningful points. And the net result of that dissonance is this rage-inducing vapidity. It's the written equivalent of the Uncanny Valley.
> Here is something that gets lost in all the excitement about AI productivity: most software engineers became engineers because they love writing code.
1) I guess I am not included in the set named "most software engineers."
2) If the title is "Software Engineer," I think I should be engineering, not coding.
This has probably been beaten to death, but I think this is the biggest disciminating question between "pro ai" and "against ai" in the software world is: "Dp you do (this) becuase you like writing code, or because you like building things for the world?"
Of course I don't think it's a binary decision.
Although I more more motivated by building things, I do somewhat miss the programmer flow state I used to get more often.
It's a different skillset and way of thinking. Engineers tend to think vertically deep on technical problems. With AI, you have to think horizontally broad and vertically up on the architectural problem. The trick is to be comfortable relegating the details to AI.
One concrete example of this realization was when I was researching how to optimize my claude code environment with agents, skills, etc. I read a lot of technical documents on how these supplemental plugins work and how to create them. After an hour of reading through all this, I realized I could just ask Claude to optimize the environment for me given the project context. So I did, and it was able to point out plugins, skills, agents that I can install or create. I gave it permission to create them and it all worked out.
This was a case of where I should not think more technically deeper, but at a more "meta" level to define the project enough for Claude to figure out how to optimize the environment. Whether that gave real gains is another question of course. But I have anecdotally observed faster results and less token usage due context caching and slightly more tools-directed prompts.
The post is right superficially. It made being an engineer harder because it took away the easy parts that anyone can do and it forces engineers to think of the hard ones.
No jobs get easier with automation - they always move a step up in abstraction level.
An accountant who was super proficient in adding numbers no longer can rely on those skills once calculator was invented.
I disagree on making it easier. I'm very capable of writing code in multiple languages but it's boring and monotonous. It's getting in the way of me building the system I have in mind. I prefer the engineering (design) to writing. If I can describe my system design to something (a junior developer or an AI) and see it come to life quickly, that's great; it lets me spend more time on designing the system, or perhaps designing more systems.
That said, there are plenty of amateurs who find coding to be approachable and system design to me daunting. For them, eliminating coding and moving the focus to system design would be a nightmare.
This is the key. I haven't found that things have become harder. The hard parts are still hard, and those have been the most important and prominent parts of my job once I reached a certain level.
Exactly. I'm a principal software engineer. My job is a lot less about writing code and a lot more about planning, designing, reviewing, and training.
However, I do wonder how we will train juniors to become seniors. Perhaps the answer is that the curriculum changes from coding and data structures to architecture and design which was typically a last minute addition in college.
Te be frank, a lot of companies don't need engineers. They need someone to do the jobs "quickly", "ASAP" and that's it. They are hiring coders masqueraded as programmers who masqueraded as engineers.
I'd say this -- if you really want to be a real engineer, you should avoid many career paths out there. Potentially ANY positions DIRECTLY facing business stakeholders is at best not a good choice, and at worst deprive your already remote chance to be a good engineer. The lower level you move into, the better, because the environment FORCES you to be a true engineer -- either you don't and fail, or you do and keep the job.
"the skills that the new engineering landscape actually requires: system design, architectural thinking, product reasoning, security awareness, and the ability to critically evaluate code they did not write."
These, surely, are the skills they always needed? Anyone who didn't have these skills was little more than a human chatgpt already, receiving prompts and simply presenting the results to someone for evaluation.
That’s a great point, but yes: a lot of devs were nothing more than a glorified LLM and during reviews were just an expensive linter. Reality is catching up to those.
There's nothing new about this pattern. When the tractor was invented, the farmer didn't get to knock off early. He just started producing 10x more. Then the tractors got bigger and more powerful, and the things you used them with got more sophisticated too and suddenly you're producing 100x more.
And yet there isn't widespread unemployment. Fewer farmers were needed so fewer people became farmers. Food became cheap and plentiful. Everyone else went on to do other things that they couldn't afford to do before. Software will do the same; we will make more software with fewer people and it will become ubiquitous to the point that people will just quickly generate whatever software they need rather than do many monotonous tasks manually.
One problem I have seen IRL is AI deployment mistakes and IMO Vibe Coders need an IT/Dev Father Figure type to avoid these simple mistakes. Here is one example:
A surgeon (no coding experience) used Claude to write a web app to track certain things about procedures he had done. He deployed the app on a web hosting provided (PHP LAMP stack). He wanted to share it with other doctors, but wasn't sure if it was 'secure' or not. He asked me to read the code and visit the site and provide my opinion.
The code was pretty reasonable. The DB schema was good. And it worked as expected. However, he routinely zipped up the entire project and placed the zip files in the web root and he had no index file. So anyone who navigated to the website saw the backups named Jan-2026.backup, etc. and could download them.
The backups contained the entire DB, all the project secrets, DB connection strings, API credentials, AWS keys, etc.
He had no idea what an 'index' file was and why that was important. Last I heard he was going to ask Claude how to secure it.
> Here is something that gets lost in all the excitement about AI productivity: most software engineers became engineers because they love writing code.
This resonates somewhat, but for a different reason. My mental model is that there are two kinds of developers, the craftsmen and the artists.
The artist considers the act of writing code their actual fulfillment. They thrive on beautifully written code. They are often attached to their code to a point where they will be hurt if someone criticizes (or even deletes) it.
The craftsman understands that code exists to serve a purpose and that is to make someone's life easier. This can be a totally non-technical customer/user that now can get their work done better. It could be another developer that benefits from using a library we wrote.
The artist hates LLMs as it takes away their work and replaces their works of beauty with generic, templatized code.
The craftsman acknowledges that LLMs are another tool in the toolbelt and using them will make them create more benefits for their customers.
One supposition I see in this and so many other articles is that using AI to generate code results in not knowing how it works. I believe that's only true for "vibe coding", not for engineers using AI to generate code. The difference is in how much you plan, design, and specify upfront.
If you give an AI a very general prompt to make an app that does X, it could build that in any imaginable way. Someone who doesn't know how these things are done wouldn't understand what way was chosen and the trade-offs involved. If they don't even look at the code, they have no idea how it works at all. This is dangerous because they are entirely dependant on the AI to make good decisions and to make any changes in the future.
Someone who practices engineering by researching, considering their options, planning and designing, and creating a specification, leaves nothing up to chance. When the prompt is detailed, the outcome is constrained to the engineer's intent. If they then review the work by seeing that it wrote what they had in mind, they know that it worked and they know that the system design matches their own design. They know how it works because they designed it and they can modify that design. They can and have read the code so they can modify it without the help of the AI.
If you know what code you want generated, reviewing it is easy - just look and see if it's what you expected. If you didn't think ahead about what the code would look like, reviewing is hard because you have to start by figuring out what the codebase even does.
This goes the same for working in small iterations rather than prompting am entire application into existence. We all know how difficult it is to review large changes and why we prefer small changes. Those same rules apply for iterations regardless of whether it was written by a person or an AI.
AI code generation can be helpful if the engineer continues acting as an engineer. It's only when someone who isn't an engineer or when an engineer abdicates their responsibilities to the AI that we end up with an unmaintainable mess. It's no different than amateurs writing scripts and spreadsheets without a full understanding of the implications of their implementation. Good software comes from good engineering, not just generating code; the code is merely the language by which we express our ideas.
I get instantly turned off by a mere whiff of AI when reading something, and consequently I refuse to foist such garbage on my fellow human beings. But by god if I read another two-line (!!!) comment with an emdash on LinkedIn, I'm going to drop a bollock.
I've unfortunately stopped reading articles before reading comments here as it's all mostly garbage now. I'm not sure what people are trying to accomplish with generating blogs aside from either clout farming or marketing for their companies.
I originally thought AI-assisted writing would help synthesize what felt like original ideas I had, that I just wanted to get out there without the laborious task of editing. I didn't expect the writing to end up feeling so incredibly tired and watered-down, but upon more reflection on how the models actually work, it's not all that surprising. Uniqueness in writing is both in the style/structure and the message, and all AI seems to do is find the local maximum of both. Lately I've found myself going back to writing things myself (not all the time, depends on the task), and wishing there was a way I could just completely eliminate the slop from certain things I look at. I worry about all our minds, and the garbage-in, garbage-out net effect of this.
Ah, just in time summarizing what we went through recently. Our "leaders" officially added these to our already nonsensical list of goals.
A. Measurably demonstrate that atleast 50% of code/tests are AI generated.
B. X% Faster delivery timelines due to improved productivity tools.
You can't expect to make a pizza in 50% less time just because you bought a faster doughmaker. Specially when you don't even know whether the dough comes out under kneaded, over kneaded or as plain lumps!
Here is something that gets lost in all the excitement about AI productivity: most software engineers became engineers because they love writing code.
I think there's a big split between those who derive meaning and enjoyment from the act of writing code or the code itself vs. those who derive it from solving problems (for which the code is often a necessary byproduct). I've worked with many across both of these groups throughout my career.
I am much more in the latter group, and the past 12mo are the most fun I've had writing software in over a decade. For those in the first group, it's easy to see how this can be an existential crisis.
It might be worth mentioning studies that show the lack of productivity gains from LLM usage. These posts take it as an unequivocal given. Management might still have the expectations that certain tasks are faster. But they aren’t always connected to reality because they’re not thinking as engineers.
AI allows you to accelerate the initial build process, but I think engineering is all about craftsmanship. Today most LLMs have poor taste and chipping away the cruft matters more than ever.
This is what hapenned 10 years ago, when machine translation entered the professional translation business. Post-editing the translation was often slower than human translating sentences from scratch. Now nearly the whole industry is post-editing machine translations, and there is more and more content that is not even post-edited.
The author introduces the term "Supervision Paradox", but IMHO this is simply one instance of the "Automation Paradox" [1], which has been haunting me since I started working in IT.
Interestingly, most jobs don't incentivize working harder or smarter, because it just leads to more work, and then burn-out.
You seem to be right. The author is pumping out one such article per day. I think I've spent more time in forming my comment than they did in generating the article. Oh well :)
What I never enjoyed was looking up the cumbersome details of a framework, a programming language or an API. It's really BORING to figure out that tool X calls paging params page and pageSize while Y offset and limit. Many other examples can be added.
For me, I feel at home in so many new programming languages and frameworks that I can really ship ideas. AI really helps with all the boring stuff.
Same here. I like bringing ideas to life; code is just a means to an end. I can now give detailed designs to an AI and let it write the hundreds of lines of code in just minutes, and with far fewer typos than I would make. It's still not perfect - I have to review it all - but if I give it a proper spec in generally creates exactly what I had in mind.
Agree, it’s made programming so much fun. The other day I wrote a C# app just because it was the best language for the job, I’ve never touched .Net in my life. Worked great, clients loved it.
I can actually build nice UIs as a traditional ML engineer (no more streamlit crap). People are using them and genuinely impressed by them
I can fly through Rust and C++ code, which used to take ages of debugging.
The main thing that is clear to me is that most of the ecosystem will likely converge toward Rust or C++ soon. Languages like Python or Ruby or even Go are just too slow and messy, why would you use them at all if you can write in Rust just as fast? I expect those languages to die off in the next several years
In writing code, as in writing poetry, the mechanical labor is 5% writing, 45% editing, and 50% reading. But the only thing that makes it yours is you.
It's an extension of pretending that developer productivity can be measured in lines of code per day, as well as the managerial blindness to the fact that code can have negative value.
> Yeah but a manager can do those things. You don't need an engineer for that.
Can they really? Engineering is about keeping the whole picture in mind so that you know which lever to push and which to not push for a certain goal. Trying until you're lucky can get you to that goal, but it's costly and not sustainable. So you need someone that can work out a model for experimentation in a less costly manner.
Judgment in this case is about deciding which path to direct the project, tradeoffs is being aware that there are other paths that are better in some aspects. And responsibility is acknowledging that a bad decision will bear a personal cost.
Everyone does the above in their own domain. But I don't think I've ever see a manager wanting to do it in the engineering domain. It's more about pushing the engineer to accept the responsibility, but denying them the power of judgment.
The scenario I'm somewhat worried about is that instead of 1 PM, 1 designer and 5 developers, there will be 1 PM, 1 designer and 1 developer. Even if tech employment stays stable or even slightly increases due to Jevons paradox, the share of software developers in tech employment will shrink.
For me, one thing that completely changed almost overnight was dealing with junior developers.
In the past, I would give them an assignment and they would take a few days to return with the implementation. I was able to see them struggling, they would learn, they would communicate and get frustrated by their own solution, then iterate.
Today, there are two kinds: 1) the ones who take a marginally smaller amount of time because they’re busy learning, testing and self reviewing, and 2) the ones who watch Twitch or Youtube videos while Claude does the job and come to me after two hours with “done, what’s next” while someone has to comb through the mess.
Leadership might see #2 and think they’re better, faster. But they are just a fucking boat anchor that drags down the whole team while providing nothing more than a shitty interface to an LLM in return.
I've always been motivated by making simple solid foundations in my code the fastest way possible.
So for me being able to have AI wrote certain things extremely fast with me just doing voice to text with my specific approach, is amazing.
I am all in on everything AI and have a discord server just for openclaw and specialized per repo assistants. It really feels like when I'm busy I can throw it an issue tracker number for things.
Then I will ssh via vs code or regular ssh which forwards my ssh key from 1password. My agents have read only repo access and I can push only when I ssh in. Super secure. Sorry for the tangent to the article but I have always loved coding now I love it even more.
> One engineer captured this shift perfectly in a widely shared essay, describing how AI transformed the engineering role from builder to reviewer.
I stopped here. Was this written by an an LLM? This sentence in particular reads exactly like the author supplied said essay as context and this sentence is the LLM's summarization of it. Nowhere is the original article linked, either, further decreasing trust. Moreover, there's an ad at the bottom for some BS "talent" platform to hire the author. This article is probably an LLM generated ad.
My trust is vacated.
This makes me feel that the SWE work/identity crisis is less important than the digital trust crisis.
I'm not sure if it's made engineering harder, but it's certainly changing what it means to be a good engineer. It's no longer just about writing code. Now it's increasingly about having good taste, making the right decisions, and sometimes just being blessed with the Midas touch.
In any case, I think we should start treating the majority of code as a commodity that will be thrown away sooner or later.
But that was a large part of it. When it was difficult to write correct, well-structured code, that was a major determinant in who would get a job as a developer - ability to design and test came second. Now that generating code is automatic, it's the rest that becomes important. That works well for those of us who could do all of those things, but hurts those whose only ability was to generate code.
AI made it so individual developers can outsource their work, not just companies. Maybe there are some lessons to be learned from companies that manage outsourced work successfully.
The role of an engineer is to produce software that probably works. If you are producing more bugs it's because you're skipping provability. AI is also really good at writing tests and doing test-driven development You can get 100% branching coverage. You use a secondary LLM to review the work and make sure everything follows best practices.
LLMs Can accelerate you if you use best practices and focus on provability and quality, but if you produce slop LLMs will help you produce slop faster.
AI learned this figure of speech from humans. Even the frequency in which it is used is copied from humans. So you can't really use it to determine if something is written by an AI or not.
LLMs might follow the frequencies of the training data in their raw form, but nobody uses raw LLMs, they use models which have been RLHFed to hell and back to bias them towards specific patterns. Then newer models were trained on the output of those RLHFed models, and further RLHFed, and so on, and so on.
In practice RLHF isn't a survey of every living humans personal style or preferences though, its purpose is to make the model more useful in the eyes of the vendor, mainly by getting cheap third-world labor to nudge the model according to the vendors instructions. You don't get a subservient, sycophantic and "safe" chat interface out of unstructured data without putting your thumb on the scale, hard.
If you think that the article is written by human or that is is unclear, please go ahead. Others here on HN also have pointed out that the author shoots out such lengthy blog posts every day. And you can also see the typical emoji AI slop here: https://www.ivanturkovic.com/services/
But I have no issue with your argumentation whatsoever, it is just that I think there is more than sufficient evidence, and you think there is not.
This section very much resonated with me, even though I still haven't tried any of the AI tools:
... most software engineers became engineers because they love writing code. Not managing code. Not reviewing code. Not supervising systems that produce code. Writing it. The act of thinking through a problem, designing a solution, and expressing it precisely in a language that makes a machine do exactly what you intended. That is what drew most of us to this profession. It is a creative act, a form of craftsmanship, and for many engineers, the most satisfying part of their day.
Actually surprised none of the other comments have picked up on this, as I don't think it's especially about AI. But the periods of my career when I've been actually writing code and solving complicated technical problems have been the most rewarding times in my life, and I'd frequently work on stuff outside work time just because I enjoyed it so much. But the other times when I was just maintaining other people's code, or working on really simple problems with cookie-cutter solutions, I get so demotivated that it's hard to even get started each day. 100%, I do this job for the challenges, not to just spend my days babysitting a fancy code generation tool.
> ...most software engineers became engineers because they love writing code. Not managing code. Not reviewing code. Not supervising systems that produce code. Writing it...
A SWE who bases their entire identity and career around only writing code is not an engineer - they are a code monkey.
The entire point of hiring a Software ENGINEER is to help translate business requirements into technical requirements, and then implement the technical requirements into a tangible feature or product.
The only reason companies buy software is because the alternative means building in-house, and for most industries software is a cost-center not a revenue generator.
I don't pay (US specific) 200K-400K TCs for code monkeys, I pay that TC for Engineers.
And this does a disservice to the large portion of SWEs and former SWEs (like me) who have been in the industry because we are customer-outcome driven (how do we use code to solve a tangible customer need) and not here to write pretty code.
You might be missing that a lot of companies are giddy that the mgmt can just vibe code stuff and there's no opportunity for engineers to be involved, (except for when it crashes?). I use AI tools and they are nice, but the mgmt are mostly not logical and need someone to sort through their bullshit.
Yeah no. Almost all companies I've chatted with - from MSPs to C-Suite of F10s - expect and demand humans-in-the-loop. I'm also on a couple boards and we've aligned on the same expectation as well.
Look, AI/ML and especially LLMs are powerful, but there does remain a degree of instability and non-determinism which will require human intervention to remediate.
That said, there is a lot of dev work in companies that is a cost-center, and those are the portions that will start getting vibe coded and deployed in product with little-to-no oversight (eg. a support portal for SMBs at an enterprise), but the equivalent feature would have already been an afterthought even without LLMs and probably given to a couple SWEs we'd be fine re-orging in a quarter anyhow.
> Yes just not driven or owned by engineers. That's what I'm seeing from company and a few peer's companies
I mean, it depends on the feature/product and how critical it is to the health of the business.
Like I mentioned in my edited comment, there is a lot of dev work in companies that is a cost-center, and those are the portions that will start getting vibe coded and deployed in product with little-to-no oversight (eg. a support portal for SMBs at an enterprise), but the equivalent feature would have already been an afterthought even without LLMs and probably given to a couple SWEs we'd be fine re-orging in a quarter anyhow because we cannot justify spending $500K-750K a year (the backend cost of 3 FT SWEs or Contractors for a company) on a customer form which nets $0 in revenue and is not directly tied with pipeline generation.
We are probably talking past each other but I am saying I see:
Leaders thinking they will basically prompt out new revenue generating features with no human engineers to "figure it out". Not cost centers, low hanging fruit, etc. No these are not giant corps like Google or whatever and likely run by morons, but it was easier when they did not think they were "empowered". There is no opportunity for engineers to "think in higher abstractions" or whatever in these cases.
> Leaders thinking they will basically prompt out new revenue generating features with no human engineers to "figure it out"
Yeah and I'm telling you as one of those leaders that most of the leaders I am meeting with know this is unrealistic and non-tech enterprises.
I think the issue is, a lot of SWEs think their work actually matters to the bottom line (and PMs and execs will massage their ego - I'm guilty of doing this as well) but in reality they don't matter because they are working in a cost-center product or feature.
Every SWE on HN should sit down and ask themselves whether or not
1. The feature they are working on directly generates revenue for their employer.
2. If it does, does it generate revenue equivalent to at least 1% of overall revenue per year.
3. Whether the cost of your team of SWEs+PMs are putting the feature/product in the red (ie. If you are 3 Eng and 1 PM working on a product who's revenue is only $500K/yr).
If all of those questions are negative, your product/feature is at risk from LLMs but was already at risk of being offshored.
I feel like there's a market out there for a weekly newsletter that summarises all the AI takes like this and collects the one meaningful snippet of insight (if any)
I still feel like I'm writing code. I tell Claude what to write and I am very specific about it. There's still tons of problems for which Claude has no particular solution and it's on me and other humans to figure out what to do. For those cases where I tell it to just go off and write a whole script that I'm not even looking at, those are throwaway / low-value cases I dont care about where previously I'd not have even taken on that particular job.
Developers will become admins. Responsible for supervising and owning the outcomes of increasingly agentic engineering outputs. Trust is the most important thing in business and it’s worth more than ever.
There's always a grain of truth in everything, but the recent article by the Redis guy (sorry for the lack of name) resonated more with me. It's correct that the load in other areas is increasing also because these tools are not there yet when it comes to for lack of a better word "good taste". I work with someone who hasn't written a line of code in a year and it shows and I'm about tired dealing with the slop. But also there's a bunch of things at work that you either did a million times already, aren't really challenging problems just annoying problems hard to solve because of all the cruft, a lot of boring manual work etc. and for this it's just an amazing help to the point I am more relaxed at work than I was previously. And when it does something that is not quite there, I can either fix it manually or tell it to fix it and it usually "gets it". Of course it it ultimately replaces me I will not be relaxed but that's a different topic.
Another little thing that resonated was a tweet that said "some will use it to learn everything and some so that they don't have to learn anything ". Of course it's not really a hard truth. It's questionable how much you can learn without really getting your hands dirty. But I do think people looking at it as a tool that helps then and/or makes them better will profit more than people looking to cut corners.
Not really, I disagree. The article did slightly touched on the real issue on why people enjoy writing code, a “craftsmanship”, yes, coding is NOT engineering, it is writing, and the people who enjoy doing it are actually writers not engineers, and I always keep mentioning that. With AI however, those writers have to be doing the engineering work: the goals, architecture design, managing blueprints, process design and refining, among many other things, and that job is not easy hence why engineers are “supposedly” paid well, AI now took the writing role, and you have to do the engineering one.
Why? Because the bottleneck was never typing code. It was always understanding the problem, making architectural decisions, debugging edge cases, and most importantly - knowing what NOT to build.
AI made me faster at producing code, but it also made me produce MORE code, which means more surface area for bugs, more maintenance burden, more complexity to reason about. The discipline of "write less code" is harder now because writing code costs almost nothing.
The engineers who thrive will be the ones who can resist the temptation to over-engineer when the marginal cost of adding complexity drops to near zero.
For me, this is a bit different. Writing code has always been the bottleneck. I get most of my joy out of solving edge cases and finding optimizations. My favorite projects are when I’m given an existing codebase with the task, “When mars and venus are opposite eachother, the code gets this weird bug that we can’t reproduce.”
When a project requires me to start from scratch, it takes me a lot longer than most other people. Once I’ve thought of the architecture, I get bored with writing the implementation.
AI has made this _a lot_ easier for me.
I think the engineers who thrive wi be the ones know when to use what tool. This has been the case before AI, AI is just another tool allowing more people to thrive.
Well if you're ever in need for a complementary mind in side projects- huh, how does one connect over HackerNews?
I stopped trying to recruit cofounders because I don’t need them anymore
I can do everything I need to do by myself plus tools at a pace no set of humans can achieve
I'm not afraid of breaking stuff because it is only a small set of users. However for my own code for my professional job no way I would go that fast because I would impact millions of users.
It is insane that companies think they can replace teams wholesale while maintaining quality.
Tech-savvy people might understand this feeling, but those who are responsible for hiring will easily proceed with another candidate that goes fast.
When push comes to shove, then, programmers will opt to have food to eat over handling technical debt generation.
The assumption is that AI will continue to improve. If we get another one or two quality jumps over the next 1-3 years, which is not totally unreasonable, AI quality might be good enough.
It’s amazing for him and it works on his iPad.
However when I tried it on my iPhone it was a broken mess. Completely unusable (not because of screen size differences).
I tried getting Claude to fix it but it couldn’t do it without changing too much of the look and feel, so I dug into the code and it was thousands of lines of absolute madness. I know from using this at work that there are things I could have done. Write tests to lock in things I like etc…
But so much of the speed up was about not caring about the specifics that once I started caring about making an actual product, I was not much faster maybe not any faster at all. The bottleneck in writing a game was never in banging out code.
Ask the AI to assess the code itself and to propose ways to gradually refactor it for better cleanliness. It can be good at that stuff, but you need to make it an explicit goal.
I find it… Amusing? That’s not quite the word. That programmers—a group notoriously for making wrong estimates of how long something will take to build—continuously and confidently spew a version of this.
And it’s not even estimating how long we ourselves would take to build something, now we’re onto estimating what an undetermined team of completely made up strangers could do. It’s bonkers. It has no basis in reality.
Yes, it is. “It would take a team 6 months” is an estimate, and I don’t see how you can argue it’s not. Even if it just said it would take them longer, that would still be an estimate.
> Claude makes me 10x faster than on my own without AI
Also an estimate.
> where 10x absolutely is a completely made up number
And by your own admission, an estimate taken from the ass that you thus cannot be certain is true. Made up perception does not equal reality.
https://metr.org/blog/2025-07-10-early-2025-ai-experienced-o...
Yes you do, you already made the argument when you pointed out the “team” size and makeup was completely unspecified, therefore the number is not an estimate, it’s just a number.
When you call it an “estimate” you are adding additional unsupported specificity to something that was explicitly stated as being hand-wavy to make an obviously rhetorical point. Are you saying you can’t understand the point being made?
My 10x is based on my experience doing projects with Claude. I also said “some” tasks, not all tasks, and I didn’t specify which tasks, and I clarified that my number is made up, which is why my number is also not an estimate of anything. There are some tasks that Claude can do 10x faster than me, and there are some tasks that it can do 100x faster than me, and there are some tasks I can do faster than Claude... for now... More importantly for me personally, Claude makes starting projects and using tech I don’t already know easier; it’s lower effort, regardless of speed.
The paper is interesting and a valid data point, but I don’t think it proves your point. I’ll respond with a few thoughts.
First, the dev’s self estimate of AI productivity speedup was +20%, even though their measured productivity was -20%. This may relate to the effort and not the speed, and it’s important to note that this is a gray zone the paper didn’t explore, and something that can be true on both sides. I can be “faster” at developing and still take the same or longer wall clock time. Measuring the time doesn’t capture how the time was spent, nor the qualities of that time.
Second, this study was done a year ago. That’s an eternity in AI land, and everyone noticed Claude and other models getting substantially better at code writing last fall, plus workflows and tooling are improving even faster than that. There’s every reason to believe the outcome of the exact same study might be different this year than it was last year.
Third, this study is explicitly biased toward large projects, and large projects are, even today, more difficult to find the productivity boosts with. I find Claude absolutely amazing at starting new projects, and absolutely terrible at working in large code bases that don’t fit in context. When I say Claude makes me 10x faster at some projects, I’m referring to something like setting up a new CRUD app when I don’t know much about setting up a database and web server backend, or writing a graphics app in Vulkan when I’ve only used OpenGL. Doing stuff like that, having Claude help me with tech stacks I don’t know, absolutely is many multiples faster than doing it on my own, and the paper link you’ve shared doesn’t address that use of AI at all.
Note specifically the paper says they are not demonstrating or claiming that “AI systems do not currently speed up many or most software developers”, and they have not demonstrated or claimed that “AI systems in the near future will not speed up developers in our exact setting”. It might be a mistake on your part to try to use this as some kind of evidence that AI isn’t speeding devs up.
This would augment actual engineer code reviews and help deal with volume.
Dunno man. Ideas alone aren't worth anything [0] and execution is everything [1], but good ideas and great execution will never go out of style regardless of how much competition is out there. I'm of the opinion that even if 10% of the population is now capable of creating a side project, there's still the same relatively-fixed amount of people capable of making a good side project, and even fewer who will see it through to a real product. Nothing has really changed in the aggregate. It's like architecture, there are always improvements in materials, tools and processes, and Claude and Codex can provide more laborers for almost free, but most people are still gonna be building uninspired McMansions instead of the Guggenheim.
[0] https://youtu.be/YYkj2yYaGtU?t=112
[1] https://youtu.be/YYkj2yYaGtU?t=160
Why do you look at it that way? Why does anyone beside you have to care about what you do?
Just build something for yourself. You will always have things you'd like to build for yourself. You will be in competition with yourself only and your target audience will be yourself.
Market forces do not apply to side-projects, because that's what people do for fun.
Just because there are chess computers, doesn't mean that no one plays chess anymore at home.
This is just a correction of something that managed to remain in an invalid state for an impressively long time.
Yes.
> The entire open source ecosystem works on this idea otherwise there would be no point in sharing and we can move to closed software.
No.
The _actual_ open source system consisted of hackers scratching their own itch and sharing the artifacts, because (it was assumed that) sharing is free. So if the work is already done and solved their problem, why not also share it as gift.
This remains unchanged.
The driving force of FOSS is not "how can I fix someone else's problem". It never has been.
Well.. maybe on HN it was different, but that's not "the open source ecosystem". And, yes, maybe some corps have gaslit naive people into believing that they must donate their lives to said corps.
If you have the time tona scratch your own itch and gift the results, it implies you have a source of income that gives you the time/lifestyle to do such a thing. You might be a tenured academic, or live in a society with a strong safety net. Or you might be able to do your day job in 1/2 the allotted time.
The problem is that a those scenarios are eroding precipitously, leaving more to seek compensation for their work output, whether it is closed or open source.
a) Almost no one but you cares and
b) Now that this has become trivial, there's no much joy in it. The struggle we had before A.I was the real joy; prompting agents for a few days and getting what you want isn't that joyful.
If you have no reaction at all, you probably weren't paying attention.
Eventually though, people _should_ recover and return after having processed the changes. So maybe the professor was still recovering at the time?
The whole side project or even private project thing doesn't just hinge on being able to produce software. There's a lot more.
In software it's the same thing. People don't really want software they want data and data transformation. But traditionally the proxy for that has been selling the software (either as a desktop app or then later as sole kind of service).
You could argue that in either case the proxy is not what people want but yet because of the difficulty of selling the "actual" thing the proxy market has flourished.
We're now inventing a new tool that will completely disrupt that market and any software business that is predicated on the complexity required to create the software to transform the data is going to get severely disrupted. Software itself will be worthless.
The value of computers since its inception was that it's capable of transforming data very, very fast and autonomously. But someone has to input that data from the real world or capture it using some device, and someone has to write the rules.
What happened is that we created a whole world of information and the rules has become very complex. Now we have multiple layers stacked vertically and multiple domains spread horizontally. At one time, ASCII was enough, now we have to deal with Unicode.
Software becoming worthless will mean that everyone has learned the rules of the systems we created and capable of creating systems with good enough quality. I'm not seeing that happens anytime soon.
When you drive down that cost you drive down the potential value of the software products. Remember that what is a cost to one party is revenue to the other party. Without revenue there cannot be profit and without revenue software has no dollar value.
If anyone can create "photoshop" with minimal cost and there are thousands of said "photoshop" apps what will be the retail sell value of those apps. Close to zero.
This same lifecycle already happened with games. Driving down the cost of producing games resulted in a proliferation of games that are mostly worthless that you can't even give away.
I do agree with you on that point.
> If anyone can create "photoshop" with minimal cost and there are thousands of said "photoshop" apps what will be the retail sell value of those apps. Close to zero.
This is the point that I cannot agree with. Not anyone can create photoshop because of the amount of knowledge you need about the data and transformations that needs to be applied to get a specific result. And then make a coherent system around it. You can create isolated function just fine, just like a lot of people knows how to build a shed with planks and nails. But even when given all the materials and tools, only a few can build a skyscraper or a mansion.
That knowledge of how to create a coherent systems that does something well is the real cost of software. Producing code isn't it.
That being said what already exists was already enough to shutter the stock prices of many software companies precisely because the fear is that their clients will just re-create the software themselves instead of buying it from someone else.
I guess we'll see how this will pan out in the next few years.
A lot of the moats are gone, but quality (and security) is in a nose dive. AI built project might be the Ikea furniture. Good for the masses, but there's still a market (much smaller) for well crafted applications and services. It's hard to say what it'll look like in a couples years though. Maybe even the crafting is eventually gone. /shrug
No one else will want this specific piece of software. But I love it.
Sure, there will be 100x the competition, but there will be also 100x the software needs. Now, if you want to get crazy rich building software, that does get tougher, but that's a good thing, I think.
Even if they were I disagree that 10x more ideas being produced means 10x more products in competition. You could leverage AI to execute but still have terrible ideas, leadership, product stewardship etc.
I think some clever people with a real and valuable insight will finally be able to turn that insight into a product. I also think the other 9 products will be get rich quick attempts by people with nothing to offer.
I think there's more opportunity to do something novel.
AI can't do it, and the humans with the skills to do it are rapidly disappearing.
> Why? Because the bottleneck was never typing code.
Were you also shipping side projects every 2 months before AI?
If not, this comment just reads like cognitive dissonance. Your core claim is that AI has enabled you to ship 7 projects in 12 months, which presumably was not something you did pre-AI, right? So the AI is helping ship projects faster?
I agree that AI is not a panacea and a skilled developer is required. I also agree that it can become a trap to produce a lot of bad code if you’re not paying attention (something a lot of companies are going to discover in 2026 IMO)
But I don’t know how you can claim AI isn’t helping you ship faster right after telling us AI is helping you ship faster.
I can guarantee you this... the story is not absolute. Depending on who you are and what you need to work on dev time could be slower, same or faster for you. BUT what we don't know is the proportion. Is it faster for 60% of people? 70%, 80%?
This is something we don't know for sure yet. But i suspect your instinct is completely wrong and that 90% of people are overall faster... much faster. I do agree that it produces more bugs and more maintenance hurdles but it is that much faster.
The thing is LLMs can bug squash too. AND they are often much faster at it then humans. My agentic set up just reads the incoming slack messages on the issue, makes a ticket, fixes the code and creates a PR in one shot.
I'm sure it also helps translate an app written for iOS into an app written for Android.
So it definitely improves performance.
Since managing dependencies is one of the major maintenance burdens in some of my projects (updating them, keeping their APIs in mind, complexity due to overgeneralization), this can help quite a lot.
See also https://www.karl.berlin/simplicity-by-llm.html for some of my thoughts regarding this.
Anything else? I'll struggle and grow as a developer, thanks. And before anyone says "but there are architecture decisions etc. so you still grow"... those existed anyways. If I have to practice, I'll practice micro AND macro skills.
The AI can help you in these tasks too, but you need to ask for the help in terms that it can help you with, and not expect it to be genuinely intelligent or to have a crystal ball. As a bonus, once you've gotten these things into the agentic context, the code itself becomes better too.
One-shotted vibe coding is an anti-pattern.
I think from time to time, it's better to ask the AI whether the codebase could be cleaned and simplified. Much better if you use different AI than what you use to make the project.
You can use it to discuss about what you should build, identify edge cases, ask you questions to force you to take decisions, etc.
As you mentioned, scope definition and constraints play a major role but ensuring that you don't just go for the first slop result but refine it pays off. It helps to have a very clear mental model of feature constraints that doesn't fall prey to scope creep.
Were you able to fairly split test?
Why will the 8th project still have those things as the bottleneck given your experience?
Also if you're not seeing any real gains in productivity, why are you using AI for your side projects and wasting tokens/money?
At work, I was told to use AI but it doesn't actually work for anything that I couldn't have handed off to a brand new undergraduate intern. So I use it for things that I don't care about then go spend twice as long rewriting what it output because it made the task longer by being wrong.
One area --and many may not like that fact-- where it can help greatly is that the cost of adding tests also drops to near zero and that doesn't work against us (because tests are typically way more localized and aren't the maintenance burden production code is). And a some us were lazy and didn't like to write too many tests. Or take generative testing / fuzzy testing: writing the proper generators or fuzzers wasn't always that trivial. Now it could become much easier.
So we may be able to use the AI slop to help us have more correct code. Same for debugging edge cases: models can totally help (I've had case as simple as a cryptic error message which I didn't recognize: passed it + the code to a LLM and it could tell me what the error was).
But yup it's a given that, as you put it, when the marginal cost of adding complexity drops to near zero, we're opening a whole new can of worms.
TFA is AI slop but fundamentally it may not be incorrect: the gigantic amount of generated sloppy code needs to be kept in check and that's where engineering is going to kick in.
That this kind of writing puts a great number of us off is not important to many who seek their fortune in this industry.
I hear the cry: "it's my own words the LLM just assisted me". Yes we have to write prompts.
I'll let an LLM update code documentation or even write a README for my project but I'll edit that to ensure it doesn't express opinions or say things like "This is designed to help make code easier to maintain" - because that's an expression of a rationale that the LLM just made up.
I use LLMs to proofread text I publish on my blog. I just shared my current prompt for that here: https://simonwillison.net/guides/agentic-engineering-pattern...
I'm not shy to admit that LLMs even from 2 years ago could communicate ideas much better than me, especially for a general audience.
It’s like everything else that AI can do - looks fine at a glance, or to the inexperienced, but collapses under scrutiny. (By your own admission you’re not a great communicator… how can you tell then?)
Thankfully we don't have to know how to write well to enjoy a well written book.
It doesn't come naturally to the more introverted type of person who cares about the object level problem and not whatever anyone else may know or doubt, I'll admit this. But slapping LLMs on it is not a great solution.
A lot of the time, the inability to express an idea clearly hints at some problem with the underlying idea, or in one's conceptualisation of that idea. Writing is a fantastic way to grapple with those issues, and iron out better and clearer iterations of ideas (or one's understanding thereof).
An LLM, on the other hand, will happily spit out a coherent piece of writing defending any nonsense idea you throw at it. Nothing is learnt, nothing is gained from such "writing" (for either the author or the audience).
It seems to bother people, perhaps since it may have been low-effort. Doesn't it not matter as long as the content is good? Otherwise, it seems to be no different than a standard low-quality post.
We should probably normalize publishing things in our native languages, and expecting the audience to run it through a translator. (I have been toying with the idea of writing everything in Esperanto (not my native language, but a favorite) and just posting links to auto-translated English versions where the translation is good enough).
EDIT: as someone with friends and family from Eastern Europe, I can tell you that the prevailing attitude is: "everything is bullshit anyway" (which, to be fair, has a lot of truth to it), and so it is no surprise that people would enthusiastically embrace a pocket-sized bullshit factory, hook it up to a fire-hose, and start spraying. We saw it with spam, and we see it now with slop. It won't stop unless the system stops rewarding it.
I don’t think there will be a point in coming to this site if it’s just going to be slop on the front page all the time.
Maybe mods should consider a tag or flag for AI generated content submissions?
Reading AI code is very pleasant. It's well annotated and consistent - how I like to read code (although not how I write code LOL). Reading language/opinions is not meant to be this way. It becomes repetitive, boring, and feels super derivative. Why would you turn the main way we communicate with each other into a soulless, tedious, chore?
I think with coding it's because I care* about what the robot is doing. But, with communication, I care about what the person is thinking in their mind, not through the interpretation of the robot. Even if the person's mind isn't as strong. At least then I can size the person up - which is the other reason understanding each other is important and ruined when you put a robot in between.
If you're talking to someone on the phone and halfway through they identify themselves as a bot, surprising you, there's a profound sense of something like betrayal. A moment ago you were having a human connection, and suddenly that vaporized. You were misled and were just talking to an unfeeling robot.
And heartfelt writing is similar. We imagine the human at the other side of the screen and we relate. And when we discover it was a bot, no matter how accurate the sentiment, that relationship vanishes.
But with math and software, it's already sterile from a human connection perspective. It's there for a different purpose. Yes, it can be beautiful, but when we read it we don't tend to build a human connection with the coder.
An interesting exception is comments. When we read the fast inverse square root code and see the "what the fuck..." comment, we instantly relate to the person writing the software. If we later learned that comment was generated by an LLM, we'd lose that connection, again.
IMHO. :)
Like look at this paragraph:
> Junior engineers have traditionally learned by doing the simpler, more task-oriented work. Fixing small bugs. Writing straightforward features. Implementing well-defined tickets. This hands-on work built the foundational understanding that eventually allowed them to take on more complex challenges.
The first sentence was enough to convey everything you needed to know, but it kept on adding words in that AI cadence. The entire post is filled with this style of writing, which, even if it is not AI, is extremely annoying to read.
Here's another example from the blog:
> Here is something that gets lost in all the excitement about AI productivity: most software engineers became engineers because they love writing code.
> Not managing code. Not reviewing code. Not supervising systems that produce code. Writing it. The act of thinking through a problem, designing a solution, and expressing it precisely in a language that makes a machine do exactly what you intended. That is what drew most of us to this profession. It is a creative act, a form of craftsmanship, and for many engineers, the most satisfying part of their day.
can just be:
> Most software engineers became engineers because they love writing code. It is a creative act, a form of craftsmanship, and for many engineers, the most satisfying part of their day.
Clarity is something that is taught in every writing class but AI generated text always seems to have this weird cadance as follows: The sound is loud. Not a whimper, not a roar, a simple sound that is very loud. And that's why... blah blah blah.
You have to care about your readers if you're writing something seriously. Throwing just a bunch of text that all mean the same thing in your writing is one of the bigger sins you can do, and that's why most people hate reading AI writing.
The part you'd like to remove ("Not managing code...") may be not required to convey the objective meaning of the sentence, but humans have emotions, too. I could have written stuff like that. To build up a bigger emotional picture.
> The act of thinking through a problem, designing a solution, and expressing it precisely in a language that makes a machine do exactly what you intended.
This sentence may not be relevant for whatever you experience to be the relevant message of the text. But it still says something the remaining paragraph does not. And also something I can relate to.
Also, as LLMs are statistical models, one has to assume that they write like this because their training data tells them to. Because humans write like this. Not when they do professional writing maybe, but when they just ramble. Not all blogs are written by professionals. I'd say most aren't. LLM training data consists mostly of humans rambling.
I also sometimes write long comments on the internet. And while I have no example to check, I feel like I do write such sentences, expanding on details to express more emotional context. Because I'm not a robot and I like writing a lot. I think it's a perfectly human thing to do. I find it sad that "writing more than absolutely needed" is now regarded as a sign of AI writing.
I would read the hell out of Joyce’s Perl 5 documentation, but only after six or seven beers.
5 sentence paragraph. First sentence is parataxis claim. Followed by 3 examples in sentence fragments, missing verbs, that familiar cadence. Then the final sentence, in this case also missing a verb.
Pure AI slop.
I hate it. I couldn't read much more after that.
Looks like something AI would say. Regardless of how it really was written
Admittedly it was so long and basic, I stopped halfway.
That's probably just default settings though - I asked it to rewrite, and most of the tell-tale signs are gone as I can see (apart from the em-dash)
https://chatgpt.com/s/t_69a46b290fb08191ad3bd93066b8cad4
1) I guess I am not included in the set named "most software engineers." 2) If the title is "Software Engineer," I think I should be engineering, not coding.
This has probably been beaten to death, but I think this is the biggest disciminating question between "pro ai" and "against ai" in the software world is: "Dp you do (this) becuase you like writing code, or because you like building things for the world?"
Of course I don't think it's a binary decision.
Although I more more motivated by building things, I do somewhat miss the programmer flow state I used to get more often.
I don't think this is true. I'm pretty sure most of them do it because it pays good salary.
One concrete example of this realization was when I was researching how to optimize my claude code environment with agents, skills, etc. I read a lot of technical documents on how these supplemental plugins work and how to create them. After an hour of reading through all this, I realized I could just ask Claude to optimize the environment for me given the project context. So I did, and it was able to point out plugins, skills, agents that I can install or create. I gave it permission to create them and it all worked out.
This was a case of where I should not think more technically deeper, but at a more "meta" level to define the project enough for Claude to figure out how to optimize the environment. Whether that gave real gains is another question of course. But I have anecdotally observed faster results and less token usage due context caching and slightly more tools-directed prompts.
No jobs get easier with automation - they always move a step up in abstraction level.
An accountant who was super proficient in adding numbers no longer can rely on those skills once calculator was invented.
That said, there are plenty of amateurs who find coding to be approachable and system design to me daunting. For them, eliminating coding and moving the focus to system design would be a nightmare.
This is the key. I haven't found that things have become harder. The hard parts are still hard, and those have been the most important and prominent parts of my job once I reached a certain level.
However, I do wonder how we will train juniors to become seniors. Perhaps the answer is that the curriculum changes from coding and data structures to architecture and design which was typically a last minute addition in college.
I'd say this -- if you really want to be a real engineer, you should avoid many career paths out there. Potentially ANY positions DIRECTLY facing business stakeholders is at best not a good choice, and at worst deprive your already remote chance to be a good engineer. The lower level you move into, the better, because the environment FORCES you to be a true engineer -- either you don't and fail, or you do and keep the job.
These, surely, are the skills they always needed? Anyone who didn't have these skills was little more than a human chatgpt already, receiving prompts and simply presenting the results to someone for evaluation.
A surgeon (no coding experience) used Claude to write a web app to track certain things about procedures he had done. He deployed the app on a web hosting provided (PHP LAMP stack). He wanted to share it with other doctors, but wasn't sure if it was 'secure' or not. He asked me to read the code and visit the site and provide my opinion.
The code was pretty reasonable. The DB schema was good. And it worked as expected. However, he routinely zipped up the entire project and placed the zip files in the web root and he had no index file. So anyone who navigated to the website saw the backups named Jan-2026.backup, etc. and could download them.
The backups contained the entire DB, all the project secrets, DB connection strings, API credentials, AWS keys, etc.
He had no idea what an 'index' file was and why that was important. Last I heard he was going to ask Claude how to secure it.
This resonates somewhat, but for a different reason. My mental model is that there are two kinds of developers, the craftsmen and the artists.
The artist considers the act of writing code their actual fulfillment. They thrive on beautifully written code. They are often attached to their code to a point where they will be hurt if someone criticizes (or even deletes) it.
The craftsman understands that code exists to serve a purpose and that is to make someone's life easier. This can be a totally non-technical customer/user that now can get their work done better. It could be another developer that benefits from using a library we wrote.
The artist hates LLMs as it takes away their work and replaces their works of beauty with generic, templatized code.
The craftsman acknowledges that LLMs are another tool in the toolbelt and using them will make them create more benefits for their customers.
If you give an AI a very general prompt to make an app that does X, it could build that in any imaginable way. Someone who doesn't know how these things are done wouldn't understand what way was chosen and the trade-offs involved. If they don't even look at the code, they have no idea how it works at all. This is dangerous because they are entirely dependant on the AI to make good decisions and to make any changes in the future.
Someone who practices engineering by researching, considering their options, planning and designing, and creating a specification, leaves nothing up to chance. When the prompt is detailed, the outcome is constrained to the engineer's intent. If they then review the work by seeing that it wrote what they had in mind, they know that it worked and they know that the system design matches their own design. They know how it works because they designed it and they can modify that design. They can and have read the code so they can modify it without the help of the AI.
If you know what code you want generated, reviewing it is easy - just look and see if it's what you expected. If you didn't think ahead about what the code would look like, reviewing is hard because you have to start by figuring out what the codebase even does.
This goes the same for working in small iterations rather than prompting am entire application into existence. We all know how difficult it is to review large changes and why we prefer small changes. Those same rules apply for iterations regardless of whether it was written by a person or an AI.
AI code generation can be helpful if the engineer continues acting as an engineer. It's only when someone who isn't an engineer or when an engineer abdicates their responsibilities to the AI that we end up with an unmaintainable mess. It's no different than amateurs writing scripts and spreadsheets without a full understanding of the implications of their implementation. Good software comes from good engineering, not just generating code; the code is merely the language by which we express our ideas.
A. Measurably demonstrate that atleast 50% of code/tests are AI generated.
B. X% Faster delivery timelines due to improved productivity tools.
You can't expect to make a pizza in 50% less time just because you bought a faster doughmaker. Specially when you don't even know whether the dough comes out under kneaded, over kneaded or as plain lumps!
I think there's a big split between those who derive meaning and enjoyment from the act of writing code or the code itself vs. those who derive it from solving problems (for which the code is often a necessary byproduct). I've worked with many across both of these groups throughout my career.
I am much more in the latter group, and the past 12mo are the most fun I've had writing software in over a decade. For those in the first group, it's easy to see how this can be an existential crisis.
Interestingly, most jobs don't incentivize working harder or smarter, because it just leads to more work, and then burn-out.
[1] https://en.wikipedia.org/wiki/Automation#Paradox_of_automati...
What I never enjoyed was looking up the cumbersome details of a framework, a programming language or an API. It's really BORING to figure out that tool X calls paging params page and pageSize while Y offset and limit. Many other examples can be added. For me, I feel at home in so many new programming languages and frameworks that I can really ship ideas. AI really helps with all the boring stuff.
AI makes using them a breeze.
I can actually build nice UIs as a traditional ML engineer (no more streamlit crap). People are using them and genuinely impressed by them
I can fly through Rust and C++ code, which used to take ages of debugging.
The main thing that is clear to me is that most of the ecosystem will likely converge toward Rust or C++ soon. Languages like Python or Ruby or even Go are just too slow and messy, why would you use them at all if you can write in Rust just as fast? I expect those languages to die off in the next several years
Maybe this is not entirely true yet, but it most likely will be in the near future.
Can they really? Engineering is about keeping the whole picture in mind so that you know which lever to push and which to not push for a certain goal. Trying until you're lucky can get you to that goal, but it's costly and not sustainable. So you need someone that can work out a model for experimentation in a less costly manner.
Judgment in this case is about deciding which path to direct the project, tradeoffs is being aware that there are other paths that are better in some aspects. And responsibility is acknowledging that a bad decision will bear a personal cost.
Everyone does the above in their own domain. But I don't think I've ever see a manager wanting to do it in the engineering domain. It's more about pushing the engineer to accept the responsibility, but denying them the power of judgment.
The scenario I'm somewhat worried about is that instead of 1 PM, 1 designer and 5 developers, there will be 1 PM, 1 designer and 1 developer. Even if tech employment stays stable or even slightly increases due to Jevons paradox, the share of software developers in tech employment will shrink.
In the past, I would give them an assignment and they would take a few days to return with the implementation. I was able to see them struggling, they would learn, they would communicate and get frustrated by their own solution, then iterate.
Today, there are two kinds: 1) the ones who take a marginally smaller amount of time because they’re busy learning, testing and self reviewing, and 2) the ones who watch Twitch or Youtube videos while Claude does the job and come to me after two hours with “done, what’s next” while someone has to comb through the mess.
Leadership might see #2 and think they’re better, faster. But they are just a fucking boat anchor that drags down the whole team while providing nothing more than a shitty interface to an LLM in return.
So for me being able to have AI wrote certain things extremely fast with me just doing voice to text with my specific approach, is amazing.
I am all in on everything AI and have a discord server just for openclaw and specialized per repo assistants. It really feels like when I'm busy I can throw it an issue tracker number for things.
Then I will ssh via vs code or regular ssh which forwards my ssh key from 1password. My agents have read only repo access and I can push only when I ssh in. Super secure. Sorry for the tangent to the article but I have always loved coding now I love it even more.
Is this still true?
I stopped here. Was this written by an an LLM? This sentence in particular reads exactly like the author supplied said essay as context and this sentence is the LLM's summarization of it. Nowhere is the original article linked, either, further decreasing trust. Moreover, there's an ad at the bottom for some BS "talent" platform to hire the author. This article is probably an LLM generated ad.
My trust is vacated.
This makes me feel that the SWE work/identity crisis is less important than the digital trust crisis.
In any case, I think we should start treating the majority of code as a commodity that will be thrown away sooner or later.
I wrote something about this here: https://chatbotkit.com/reflections/most-code-deserves-to-die - it was inspired by another conversation on HN.
It never was
That's different than saying a lot of people *believed* writing code was the hardest/most important part.
LLMs Can accelerate you if you use best practices and focus on provability and quality, but if you produce slop LLMs will help you produce slop faster.
> That is not an upgrade. That is a career identity crisis.
This is not X. It is Y.
> The trap is ...
> This gap matters ...
> This is not empowerment ...
> This is not a minor adjustment...
Your typical AI slop rhetorical phrasing.
Phrases like: "identity crisis", "burnout machine", "supervision paradox", "acceleration trap", "workload creep"
These sound analytical but are lightly defined. They function as named concepts without rigorous definition or empirical grounding.
There might be some good arguments in the article, but AI slop remains AI slop.
> AI is an in-context learner, not a standards enforcer.
> The AI is not judging your code. It is learning from it.
> Speed without structure is not speed. It is borrowed time.
> This is not about premature optimization or over-engineering. It is about giving the AI the patterns it needs to work effectively on your behalf.
> This is not a theoretical distinction. It is the single most important practical reality of working with AI coding tools in 2026.
Its not this, its that.
> But here is the part nobody wants to hear: the reverse is equally true.
> The result was transformative.
> Here is why.
If you want I can provide N=3 with the same AI pattern and phrases again.
But I have no issue with your argumentation whatsoever, it is just that I think there is more than sufficient evidence, and you think there is not.
... most software engineers became engineers because they love writing code. Not managing code. Not reviewing code. Not supervising systems that produce code. Writing it. The act of thinking through a problem, designing a solution, and expressing it precisely in a language that makes a machine do exactly what you intended. That is what drew most of us to this profession. It is a creative act, a form of craftsmanship, and for many engineers, the most satisfying part of their day.
Actually surprised none of the other comments have picked up on this, as I don't think it's especially about AI. But the periods of my career when I've been actually writing code and solving complicated technical problems have been the most rewarding times in my life, and I'd frequently work on stuff outside work time just because I enjoyed it so much. But the other times when I was just maintaining other people's code, or working on really simple problems with cookie-cutter solutions, I get so demotivated that it's hard to even get started each day. 100%, I do this job for the challenges, not to just spend my days babysitting a fancy code generation tool.
A SWE who bases their entire identity and career around only writing code is not an engineer - they are a code monkey.
The entire point of hiring a Software ENGINEER is to help translate business requirements into technical requirements, and then implement the technical requirements into a tangible feature or product.
The only reason companies buy software is because the alternative means building in-house, and for most industries software is a cost-center not a revenue generator.
I don't pay (US specific) 200K-400K TCs for code monkeys, I pay that TC for Engineers.
And this does a disservice to the large portion of SWEs and former SWEs (like me) who have been in the industry because we are customer-outcome driven (how do we use code to solve a tangible customer need) and not here to write pretty code.
Look, AI/ML and especially LLMs are powerful, but there does remain a degree of instability and non-determinism which will require human intervention to remediate.
That said, there is a lot of dev work in companies that is a cost-center, and those are the portions that will start getting vibe coded and deployed in product with little-to-no oversight (eg. a support portal for SMBs at an enterprise), but the equivalent feature would have already been an afterthought even without LLMs and probably given to a couple SWEs we'd be fine re-orging in a quarter anyhow.
> but there does remain a degree of instability and non-determinism which will require human intervention to remediate.
I agree.
I mean, it depends on the feature/product and how critical it is to the health of the business.
Like I mentioned in my edited comment, there is a lot of dev work in companies that is a cost-center, and those are the portions that will start getting vibe coded and deployed in product with little-to-no oversight (eg. a support portal for SMBs at an enterprise), but the equivalent feature would have already been an afterthought even without LLMs and probably given to a couple SWEs we'd be fine re-orging in a quarter anyhow because we cannot justify spending $500K-750K a year (the backend cost of 3 FT SWEs or Contractors for a company) on a customer form which nets $0 in revenue and is not directly tied with pipeline generation.
Leaders thinking they will basically prompt out new revenue generating features with no human engineers to "figure it out". Not cost centers, low hanging fruit, etc. No these are not giant corps like Google or whatever and likely run by morons, but it was easier when they did not think they were "empowered". There is no opportunity for engineers to "think in higher abstractions" or whatever in these cases.
Yeah and I'm telling you as one of those leaders that most of the leaders I am meeting with know this is unrealistic and non-tech enterprises.
I think the issue is, a lot of SWEs think their work actually matters to the bottom line (and PMs and execs will massage their ego - I'm guilty of doing this as well) but in reality they don't matter because they are working in a cost-center product or feature.
Every SWE on HN should sit down and ask themselves whether or not
1. The feature they are working on directly generates revenue for their employer.
2. If it does, does it generate revenue equivalent to at least 1% of overall revenue per year.
3. Whether the cost of your team of SWEs+PMs are putting the feature/product in the red (ie. If you are 3 Eng and 1 PM working on a product who's revenue is only $500K/yr).
If all of those questions are negative, your product/feature is at risk from LLMs but was already at risk of being offshored.
THE MARKET WILL FILL THAT VOID
IT DOES NOT MAKE IT TRUE
it's all so fucking tiresome
Also, check out the dude's linkedin: https://www.linkedin.com/in/ivanturkovic/
Another little thing that resonated was a tweet that said "some will use it to learn everything and some so that they don't have to learn anything ". Of course it's not really a hard truth. It's questionable how much you can learn without really getting your hands dirty. But I do think people looking at it as a tool that helps then and/or makes them better will profit more than people looking to cut corners.