This is pretty recent - the survey they ran (99 respondents) was August 18 to September 23 2025 and the field observations (watching developers for 45 minute then a 30 minute interview, 13 participants) were August 1 to October 3.
The models were mostly GPT-5 and Claude Sonnet 4. The study was too early to catch the 5.x Codex or Claude 4.5 models (bar one mention of Sonnet 4.5.)
This is notable because a lot of academic papers take 6-12 months to come out, by which time the LLM space has often moved on by an entire model generation.
Depends on the participants. If they're cutting-edge LLM users then yes, I think so. If they continue to use LLMs like they would have back in the first half of 2025 I'm not sure if a difference would be noticeable.
I'm not remotely cutting edge (just switched from Cursor to Codex CLI, have no fancy tooling infrastructure, am not even vaguely considering git worktrees as a means of working), but Opus 4.5 and 5.2 Codex are both so clearly more competent than previous models that I've started just telling them to do high-level things rather than trying to break things down and give them subtasks.
If people are really set in their ways, maybe they won't try anything beyond what old models can do, and won't notice a difference, but who's had time to get set in their ways with this stuff?
I mostly agree, but today, Opus 4.5 via Claude code did something pretty dumb stuff in my codebase— N queries where one would do, deep array comparison where a reference equality check would suffice, very complex web of nested conditionals which a competent developer would have never written, some edge cases where the backend endpoints didn’t properly verify user permissions before overwriting data, etc.
It’s still hit or miss. The product “worked” when I tested it as a black box, but the code had a lot of rot in it already.
Maybe that stuff no longer matters. Maybe it does. Time will tell.
I don't think they generally one-shot the tasks; but they do them well enough that you can review the diff and make requests for changes and have it succeed in a good outcome more quickly than if you were spoon-feeding it little tasks and checking them as you go (as you used to have to do).
I have had a lot of success lately when working with Opus 4.5 using both the Beads task tracking system and the array of skills under the umbrella of Bad Dave's Robot Army. I don't have a link handy, but you should be able to find it on GitHub. I use the specialized skills for different review tasks (like Architecture Review, Performance Review, Security Review, etc.) on every completed task in addition to my own manual review, and I find that that helps to keep things from getting out of hand.
As someone who’s responsible for some very clean codebases and some codebases that grew over many years, warts and all, I always wonder if being subjected to large amounts of not-exactly-wonderful code has the same effect on an LLM that it arguably also has on human developers (myself included occasionally): that they subconsciously lower their normally high bar for quality a bit, as in „well there‘s quite some smells here, let’s go a bit with the flow and not overdo the quality“.
Coding agents and much better models. Claude Code or Codex CLI plus Claude Opus 4.5 or GPT 5.2 Codex.
The latest models and harnesses can crunch on difficult problems for hours at a time and get to working solutions. Nothing could do that back in ~March.
Every single example you gave is in a hobby project territory. Relatively self-contained, maintainable by 3-4 devs max, within 1k-10k lines of code. I've been successfully using coding agents to create such projects for the past year and it's great, I love it.
However, lots of us here work on codebases that are 100x, 1000x the size of these projects you and Karpathy are talking about. Years of domain specific code. From personal experience, coding agents simply don't work at that scale the same way they do for hobby projects. Over the past year or two, I did not see any significant improvement from any of the newest models.
Building a slightly bigger hobby project is not even close to making these agents work at industrial scale.
That’s right, but it also hints at a solution: split big code bases into parts that are roughly the size of a big hobby project. You’ll need to write some docs to be effective at it, which also helps agents. CICD means continuous integration continuous documentation now.
Most of the stuff I'm talking about here came out in November. There hasn't been much time for professional teams to build new things with it yet, especially given the holidays!
I was going back and looking at timelines, and was shocked to realize that Claude Code and Cursor's default-to-agentic-mode changes both came out in late February. Essentially the entire history of "mainstream" agentic coding is ten months old.
(This helps me understand better the people who are confused/annoyed/dismissive about it, because I remember how dismissive people were about Node, about Docker, about Postgres, about Linux when those things were new too. So many arguments where people would passionately talk about all those things were irredeemably stupid and only suitable for toy/hobby projects.)
It feels like we're doing another lift to a higher level of abstraction. Whereas we had "automatic programming" and "high level programming languages" free us from assembly, where higher level abstractions could be represented without the author having to know or care about the assembly (and it took decades for the switch to happen), we now once again get pulled up another layer.
We're in the midst of another abstraction level becoming the working layer - and that's not a small layer jump but a jump to a completely different plane. And I think once again, we'll benefit from getting tools that help us specify the high level concepts we intend, and ways to enforce that the generated code is correct - not necessarily fast or efficient but at least correct - same as compilers do. And this lift is happening on a much more accelerated timeline.
The problem of ensuring correctness of the generated code across all the layers we're now skipping is going to be the crux of how we manage to leverage LLM/agentic coding.
The title is doing a lot of work here. What resonated with me is the shift from “writing code” to “steering systems” rather than the hype framing. Senior devs already spend more time constraining, reviewing, and shaping outcomes than typing syntax. AI just makes that explicit. The real skill gap isn’t prompt cleverness, it’s knowing when the agent is confidently wrong and how to fence it in with tests, architecture, and invariants. That part doesn’t scale magically.
I think there are two groups of people emerging. deep / fast / craft-and-decomposition-loving vs black box / outcome-only.
I've seen people unable to work at average speed on small features suddenly reach above average output through a llm cli and I could sense the pride in them. Which is at odds with my experience of work.. I love to dig down, know a lot, model and find abstractions on my own. There a llm will 1) not understand how my brain work 2) produce something workable but that requires me to stretch mentally.. and most of the time I leave numb. In the last month I've seen many people expressing similar views.
ps: thanks everybody for the answers, interesting to read your pov
I get what you're saying, but I would say that this does not match my own experience. For me, prior to the agentic coding era, the problem was always that I had way more ideas for features, tools, or projects than I had the capacity to build when I had to confront the work of building everything by hand, also dealing with the inevitable difficulties in procrastination and getting started.
I am a very above-average engineer when it comes to speed at completing work well, whether that's typing speed or comprehension speed, and still these tools have felt like giving me a jetpack for my mind. I can get things done in weeks that would have taken me months before, and that opens up space to consider new areas that I wouldn't have even bothered exploring before because I would not have had the time to execute on them well.
The sibling comments (from remich and sanufar) match my experience.
1. I do love getting into the details of code, but I don't mind having an LLM handle boilerplate.
2. There isn't a binary between having an LLM generate all the code and writing it all myself.
3. I still do most of the design work because LLMs often make questionable design decisions.
4. Sometimes I simply want a program to solve a problem (outcome-focused) over a project to work on (craft-focused). Sometimes I need a small program in order to focus on the larger project, and being able to delegate that work has made it more enjoyable.
> I do love getting into the details of code, but I don't mind having an LLM handle boilerplate.
My usual thought is that boilerplate tells me, by existing, where the system is most flawed.
I do like the idea of having a tool that quickly patches the problem while also forcing me to think about its presence.
> There isn't a binary between having an LLM generate all the code and writing it all myself. I still do most of the design work because LLMs often make questionable design decisions.
One workflow that makes sense to me is to have the LLM commit on a branch; fix simple issues instead of trying to make it work (with all the worry of context poisoning); refactor on the same branch; merge; and then repeat for the next feature — starting more or less from scratch except for the agent config (CLAUDE.md etc.). Does that sound about right? Maybe you do something less formal?
> Sometimes I simply want a program to solve a purpose (outcome-focused) over a project to work on (craft-focused). Sometimes I need a small program in order to focus on the larger project, and being able to delegate that work has made it more enjoyable.
I think for me, the difference really comes down to how much ownership I want to take in regards to the project. If it’s something like a custom kernel that I’m building, the real fun is in reading through docs, learning about systems, and trying to craft the perfect abstractions; but if it’s wiring up a simple pipeline that sends me a text whenever my bus arrives, I’m happy to let an LLM crank that out for me.
I’ve realized that a lot of my coding is on this personal satisfaction vs utility matrix and llms let me focus a lot more energy onto high satisfaction projects
> deep / fast / craft-and-decomposition-loving vs black box / outcome-only
As a (self-reported) craft-and-decomposition lover, I wouldn't call the process "fast".
Certainly it's much faster than if I were trying to take the same approach without the same skills; and certainly I could slow it down with over-engineering. (And "deep" absolutely fits.) But the people I've known that I'd characterize as strongly "outcome-only", were certainly capable of sustaining some pretty high delta-LoC per day.
That's kind of the point here. Once a dev reached a certain level, they often weren't doing much "relaxing code typing" anyways before the AI movement. I don't find it to be much different than being a tech lead, architect, or similar role.
As a former tech lead and now staff engineer, I definitely agree with this. I read a blog post a couple of months ago that theorized that the people that would adopt these technologies the best were people in the exact roles that you describe. I think because we were already used to having to rely on other people to execute on our plans and ideas because they were simply too big to accomplish by ourselves. Now that we have agents to do these things, it's not really all that different - although it is a different management style working around their limitations.
Exactly. I've been a tech lead, have led large, cross-org projects, been an engineering manager, and similar roles. For years, when mentoring upcoming developers what I always to be the most challenging transition was the inflection point between "I deliver most of my value by coding" to "I deliver most of my value by empowering other people to deliver". I think that's what we're seeing here. People who have made this transition are already used to working this way. Both versions have their own quirks and challenges, but at a high level it abstracts.
LLMs are just a programming language/compiler/REPL, though, so there is nothing out of the ordinary for developers. Except what is different is the painfully slow compile time to code ratio. You write code for a few minutes... and then wait. Then spend a few more minutes writing code... and then wait. That is where the exhaustion comes from.
At least in the olden days[1] you could write code for days before compiling, which reduced the pain. Long compilation times has always been awful, but it is less frustrating when you could defer it until the next blue moon. LLMs don't (yet) seem to be able to handle that. If you feed them more than small amounts of code at a time and they quickly go off the rails.
With that said, while you could write large amounts of code and defer it until the next blue moon, it is a skill to be able to do that. Even in C++, juniors seem to like to write a few lines of code and then turn to compiling the results to make sure they are on the right track. I expect that is the group of people who is most feeling at home with LLMs. Spending a few minutes writing code and then waiting on compilation isn't abnormal for them. They don't know any different!
Ya know, I have to admit feeling something like this. Normally, the amount of stuff I put together in a work day offers a sense of completion or even a bit of a dopamine bump because of a "job well done". With this recent work I've been doing, it's instead felt like I've been spending a multiplier more energy communicating intent instead of doing the work myself; that communication seems to be making me more tired than the work itself. Similar?
You’re possibly not entering into the flow state anymore.
Flow is effortless. and it is rejuvenating.
I believe:
While communication can be satisfying, it’s not as rejuvenating as resting in our own Being and simply allowing the action to unfold without mental contraction.
Flow states.
When the right level of challenge and capability align and you become intimate with the problem. The boundaries of me and the problem dissolve and creativity springs forth. Emerging satisfied. Nourished.
I forget where I saw this (a Medium post, somewhere) but someone summed this up as "I didn't sign up for this just to be a tech priest for the machine god".
Someone commented yesterday that managers and other higher-ups are "already ok with non-deterministic outputs", because that's what engineers give them.
As a manager/tech-lead, I've kind of been a tech priest for some time.
This is why I think LLMs will make us all a LOT smarter. Raw code made it so we stopped heavily thinking in between but now it's just 100% the most intense thought processes all day long.
Nah, I don’t miss at all typing all the tests, CLIs, and APIs I’ve created hundreds of times before. I dunno if I it’s because I do ML stuff, but it’s almost all “think a lot about something, do some math, and and then type thousands of lines of the same stuff around the interesting work.”
It's difficult to steer complex systems correctly, because no one has a complete picture of the end goal at the outset. That's why waterfall fails. Writing code agentically means you have to go out of your way to think deeply about what you're building, because it won't be forced on you by the act of writing code. If your requirements are complex, they might actually be a hindrance because you're going have to learn those lessons from failed iterations instead of avoiding them preemptively.
The stereotype that writing code is for junior developers needs to die. Some devs are hired with lofty titles specifically for their programming aptitude and esoteric systems knowlege, not to play implementation telephone with inexperienced devs.
I don't think that anyone actually believes that writing code is only for junior developers. That seems to be a significant exaggeration at the very least. However, it is definitely true that most organizations of this size are hiring people into technical lead, staff engineer, or principal engineer roles are hiring those people not only for their individual expertise, or ability to apply that expertise themselves, but also for their ability to use that expertise as a force multiplier to make other less experienced people better at the craft.
> Takeaway 3c: Experienced developers disagree about using agents for software
planning and design. Some avoided agents out of concern over the importance of
design, while others embraced back-and-forth design with an AI.
Im in the back-and-forth camp. I expect a lot of interesting UX to develop here. I built https://github.com/backnotprop/plannotator over the weekend to give me a better way to review & collaborate around plans - all while natively integrated into the coding agent harness.
The new layer of abstraction is tests. Mostly end-to-end and integration tests. It describes the important constraints to the agents, essentially long lived context.
So essentially what this means is a declarative programming system of overall system behavior.
we've never seen a profession drive themselves so aggressively to irrelevance. software engineering will always exist, but it's amazing the pace to which pressure against the profession is rising. 2026 will be a very happy new year indeed for those paying the salaries. :)
We've been giving our work away to each other for free as open source to help improve each other's productivity for 30+ years now and that's only made our profession more valuable.
I see little proof open source has resulted in higher wages and not the fact that everything is being digitized and the subsequent demand for such people to assist in such.
I'm not sure how I can prove it, but ~25 years ago building software without open source sucked. You had to build everything from scratch! It took months to get even the most basic things up and running.
I think open source is the single most important productivity boost to our industry that's ever existed. Automated testing is a close second.
Google, Facebook, many others would not have existed without open source to build on.
And those giants and others like them that were enabled by open source employed a TON of people, at competitive rates that greatly increased our salaries.
Agentic coding is just another rhyme of 25 y/o frenzy of "let's outsource everything to India." The new generation thinks this time is really special with us. Let's check again in 25 years
Being productive with VB6 (final release in 1998) is very far from being productive with modern stacks. Apples and orangutans.
Will I grant advantages to certain aspects of VB6? Sure. Could some lessons be applicable today? Probably. But just like historians say, don't make the mistake of ignoring context when you compare things from different eras.
P.S. Search for "most dreaded legacy languages" and you'll see a lot of discussion on VB6, much of it unflattering. I don't have a dog in that fight -- I can't tell if it is more on polarizing end of things (some love, some hate) or more on the "universally loathed" side.
I'm too young to have used VB in the workforce, but I did use it in school, and honestly off that alone I'm inclined to agree.
I've seen VB namedropped frequently, but I feel like I've yet to see a proper discussion of why it seems like nothing can match its productivity and ease of use for simple desktop apps. Like, what even is the modern approach for a simple GUI program? Is Electron really the best we can do?
MS Access is another retro classic of sorts that, despite having a lot of flaws, it seems like nothing has risen to fill its niche other than SaaS webapps like airtable.
You can add Macromedia Flash to that list - nothing has really replaced it, and as a result the world no longer has an approachable tool for building interactive animations.
Indeed it did; I remember those times. All else being equal I still think SWE salaries on average would of been higher if we kept it like that given basic economics - there would of been a lot less people capable of doing it but the high ROI automation opportunities would of still been there. The fact that "it sucked" usually creates more scarcity on the supply side; which all being equal means higher wages and in our capitalist society - status. Other professions that are older as to the parent comment already know this and don't see SWE as very "street smart" disrupting themselves. I've seen articles recently like "at least we aren't in coding" from law, accounting, etc an an anecdote to this.
With AI at least locally I'm seeing the opposite now - less hiring, less wage pressure and in social circles a lot less status when I mention I'm a SWE (almost sympathy for my lot vs respect only 5 years ago). While I don't care for the status aspect, although I do care for my ability to earn money, some do.
At least locally inflation adjusted in my city SWE wages bought more and were higher in general compared to others in the 90's-2000's than on wards (ex big tech). Partly because this difficulty and low level knowledge meant only very skilled people could participate.
I mean, this seems like a pretty big thing to leave out, no? That's where all the crazy high salaries were!
Also, there are still legacy places that more or less build software like it's 1999. I get the impression that embedded, automotive, and such still rely a lot on proprietary tools, finicky manual processes, low level languages (obviously), etc. But those are notorious for being annoying and not very well paid.
I'm talking about what I perceive to be the median salary/conditions with big tech being only a part of that. My point is more that I remember back in that period good salaries could be had outside big tech too even in the boring standard companies that you state. I remember banks, insurance, etc paying very well for example compared to today for an SWE/tech worker - the good opportunities seemed more distributed. For example I've seen contract rates for some of the people we hire haven't really changed for 10 years for developers. Now at best they are on par with other professional white collar workers; and the competition seems fiercer (e.g. 5 interviews for a similar salary with leetcode games rather than experienced based interviews).
Making software easier and more abstract has allowed less technical people into the profession, allowed easier outsourcing, meant more competition/interview prep to filter out people (even if the skills are not used in the job at all), more material for AI to train on, etc. To the parent comment's point I don't think it has boosted salaries and/or conditions on average for the SWE - in the long run (10 years +) it could be argued that economically the opposite has occurred.
Monopolizing the work doesn't work unless you have the power to suppress anyone else joining the competition, i.e. "certified developers only".
Otherwise people would have realized they can charge 3x as much by being 5x as productive with better tools while you're writing your code in notepad for maximum ROI, and you would have either adjusted or gone out of business.
Increased productivity isn't a choice, it's a result of competition. And that's a good thing overall, even if it sucks for some developers who now have to actually work for the first time in decades. But it's good for society at large, because more things can be done.
Sure - I agree with that, and I agree its good for society but as you state probably not as good for the SWE who has to work harder for the same which was my point and I think you agree. Other professions have done what you have stated (i.e. certification) and seen higher wages than otherwise which also proves my point. They see this as the "street smart" thing to do, and generally society respects them for it putting their profession on a higher pedestal as a result. People respect people who take care of themselves first generally I find as well. Personally I think there should be a balance between the two (i.e. a fair go for all parties; a fair day's work with some job security over a standard career lifetime but not extortionary).
Also your notion of "better tools" may of not happened, or happened more slowly without open source, AI, etc which would of meant higher salaries for longer most probably. That's where I disagree with the parent poster's claim of higher salaries - AI seems to be a great recent example of "better tools" disrupting the premium SWE's enjoy rather than improving their salaries. Whether that's fair or not is a different debate.
I was just doubting the notion of the parent comment that "open source software" and "automated testing" create higher salaries. Usually efficiency economically (some exceptional cases) creates lower salaries for the people who are made more efficient all else being equal - and the value shifts from them to either consumers or employers.
I don't think that's certain. I'm hoping for a Jevons paradox situation where AI drives down the cost of producing software to the point that companies that previously weren't in the market for custom software start hiring software engineers. I think we could see demand go up.
This makes sense. Imagine PHP or NodeJS without a framework, or front end development without React. Your projects would take much longer to build. The time saved with the open source frameworks and libraries is more than what an AI agent can save you.
> we've never seen a profession drive themselves so aggressively to irrelevance.
Should we be trying to put the genie back in the bottle? If not, what exactly are you suggesting?
Even if we all agreed to stop using AI tools today, what about the rest of world? Will everybody agree to stop using it? Do you think that is even a remote possibility?
Does the rest of the world want to make money in a way not involving digging ditches? I feel like people from developing countries that spend 18 hours a day studying, giving their entire childhood to some standardized test, may not want yo be rewarded with no job prospects. Maybe that’s a crazy position.
Also it really baffles me how many are actually in on the hype train. Its a lot more than the crypto bros back in the day. Good thing AI still cant reason and innovate stuff. Also leaking credentials is a felony in my country so I also wont ever attach it to my codebases.
I think the issue is folks talk past each other. People who find coding agents useful or enjoyable are labeled “on the hype train” and folks for which coding agents don’t work for them or their workflow are considered luddites. There are an incredible number of contradicting claims and predictions out there as well, and I believe what we see is folks projecting their reaction to some amalgamation of them onto others. I see a lot of “they” language, and a lot of viral articles about business leadership “shoving AI down our throats” and it becomes a divisive issue like American political scene with really no one having a real conversation
I think the reason for the varying claims and predictions is because developers have wildly different standards for what constitutes working code. For the developers with a lower threshold, AI is like crack to them because gen ai's output is similar to what they would produce, and it really is a 10x speedup. For others, especially those who have to fix and maintain that code, it's more like a 10x slowdown.
Hence why you have in the same thread, some developer who claims that Claude writes 99% of their code and another developer who finds it totally useless. And of course others who are somewhere in the middle.
Have you considered that it's a bit dismissive to assume that developers who find use out of AI tools necessarily approve of worse code than you do, or have lower standards?
It's fine to be a skeptic. Or to have tried out these tools and found that they do not work well for your particular use case at this moment in time. But you shouldn't assume that people who do get value out of them are not as good at the job as you are, or are dumber than you are, or slower than you are. That's just not a good practice and is also rude.
I never said anything about being worse, dumber, and definitely not slower. And keep in mind worse is subjective - if something doesn't require edge case handling or correctness, bugs can be tolerated etc, then something with those properties isn't worse is it?
I'm just saying that since there is such a wide range of experiences with the same tools, it's probably likely that developers vary on their evaluations of the output.
Okay, I certainly agree with you that different use cases can dictate different outcomes when using AI tooling. I would just encourage everyone who thinks similar to you to be cautious about assuming that someone who experiences a different result with these tools is less skilled or dealing with a less difficult use case - like one that has no edge cases or has greater tolerance for bugs. It's possible that this is the case, but it is just as possible that they have found a way to work with these tools that produces excellent output.
There's also the effect of different models. Until the most recent models, especially for concise algorithms, I felt it was still easier to sometimes do it myself (i.e. a good algo can be concise/more concise than a lossy prompt) and leave the "expansion/repetitive" boilerplate code to the LLM. At least for me the latest models do feel like a "step change" in that the problems can be bigger and/or require less supervision on each problem depending on the tradeoff you want.
Its all a hype train though. People still believe in the AI gonna bring utopia bullshit while the current infra is being built on debt. The only reason it still exists is that all these AI companies believe in some kind of revenue outside of subscriptions. So its all about:
Owning the infrastructure and enshittify (ads) once enough products are based on AI.
.env files are used to develop as well, for some things like PayPal u dont have to change the credentials, you just enable sandbox mode. If I had some LLM attached to my codebase, it would be able to read those credentials from the .env file.
This has nothing to do with deployment. I never talked about deployment.
That’s not how LLMs work, it’s part of the reinforcement learning or SFT dataset, data labelers would have written or generated tons of examples using this and other patterns (all the emoji READMEs for example) that the models emulate. The early ones had very formulaic essay style outputs that always ended with “in conclusion”, lots of the same kind of bullet lists, and a love of adjectives and delving, all of which were intentionally trained in. It’s more subtle now but it’s still there.
Maybe I was being imprecise, but I’m not sure what you mean by “not how LLMs work” - discovering patterns of how humans write is exactly the signal they are trained against. Either explicitly curated like SFT or coaxed out during RLHF, no?
It could even have been picked up in pretraining and then rewarded during rlhf when the output domain was being refined; I haven’t used enough LLMs before post training to know what step it usually becomes noticeable.
I often tell people that agentic programming tools are the best thing since cscope. The last 6 months I have not used cscope even once after decades of using it nearly daily.
You know what. After seeing all these articles about AI/LLM for these past 4 years, about how they are going to replace me as software developers and about how I am not productive enough without using 5 agents and being a project manager.
I. Don't. Care.
I don't even care about those debates outside. Debates about do LLM work and replace programmers? Say they do, ok so what?
I simply have too much fun programming. I am just a mere fullstack business line programmer, generic random replaceable dude, you can find me dime a dozen.
I do use LLM as Stack Overflow/docs replacement, but I always code by hand all my code.
If you want to replace me, replace me. I'll go to companies that need me. If there are no companies that need my skill, fine, then I'll just do this as a hobby, and probably flip burgers outside to make a living.
I don't care about your LLM, I don't care about your agent, I probably don't even care about the job prospects for that matter if I have to be forced to use tools that I don't like and to use workflows I don't like. You can go ahead find others who are willing to do it for you.
As for me, I simply have too much fun programming. Now if you excuse me, I need to go have fun.
I simply will not spend my life begging and coaxing a machine to output working code. If that is what becomes of this profession, I will just do something else :)
Until you realize you're just begging and coaxing a human to better beg and coax a machine to output working code - when you could just beg and coax the machine yourself.
It would definitely be the profession if we stopped developing things today. Think about the idea of coding agents 2 years ago, I personally found them very unrealistic and am now coding exclusively with them despite them being either a neutral or net negative to my development time simply because I see the writing on the wall that in 6 mos to a year they will probably be a huge net positive and in 2-3 years the dismissive attitude towards adoption will start to look kind of silly (no offense). To me we are _just_ at the inflection point where using and not using coding agents are both totally sensible decisions.
Hear hear. I didn't spend half my life getting an education, competing in the corporate crab bucket, retraining and upskilling just to turn into a robot babysitter.
I hear you but I feel like you (and really others like you, in mass) should not be so passive about your replacement. For most programmers, simply flipping burgers for money to enjoy programming a few hours a week is not going to work. Making a living is a thing. If you are reduced to having to flip burgers that means the economy will gave collapsed and there won’t be any magic Elon UBI money to save us.
not sure why so many people feel like factoring fun into what job you want to take is so unthinkable, or that it's just a false dichotomy between the ideal job and unemployment
it's a trade-off; you need a job but you typically interview at several places, collect offers, and weigh them according to various criteria. all the pro-fun posters are saying is that "enjoy the job" is a very highly ranked criterion for us.
Why? It is a matter of values. Fun can be a driving force just like money and stability is. It is simply a matter of your values (and your sacrifices).
Like I said, I am just a generic replaceable dime a dozen programmer dude.
This is a qualitative methods paper, so statistical significance is not relevant. The rough qualitative equivalent would instead be "data saturation" (responses generally look like ones you've received already) and "thematic saturation" (you've likely found all the themes you will find through this method of data collection). There's an intuitive quality to determining the number of responses needed based on the topic and research questions, but this looks to me like they have achieved sufficient thematic saturation based on the results.
Page 2:
We define agentic tools or agents as AI tools integrated into an IDE or a terminal that can manipulate the code directly (i.e., excluding web-based chat interfaces)
Idk, I still mostly avoid using it and if I do, I just copy and paste shit into the Claude web version. I wont ever manage agents as that sounds just as complicated as coding shit myself.
It's not complicated at all. You don't "manage agents". You just type your prompt into an terminal application that can update files, read your docs and run your tests.
As with every new tech there's a hell of a lot of noise (plugins, skills, hooks, MCP, LSP - to quote Kaparthy) but most of it can just be disregarded. No one is "behind" - it's all very easy to use.
This is pretty recent - the survey they ran (99 respondents) was August 18 to September 23 2025 and the field observations (watching developers for 45 minute then a 30 minute interview, 13 participants) were August 1 to October 3.
The models were mostly GPT-5 and Claude Sonnet 4. The study was too early to catch the 5.x Codex or Claude 4.5 models (bar one mention of Sonnet 4.5.)
This is notable because a lot of academic papers take 6-12 months to come out, by which time the LLM space has often moved on by an entire model generation.
I knew in October the game had changed. Thanks for keeping us in the know.
Herbalife-5.5 is more powerful than Herbalife-4.5, increases your metabolism by 30% and improves your gut biome by 10%.
Thanks Simon - always quick on the draw.
Off your intuition, do you think the same study with Codex 5.2 and Opus 4.5 would see even better results?
Depends on the participants. If they're cutting-edge LLM users then yes, I think so. If they continue to use LLMs like they would have back in the first half of 2025 I'm not sure if a difference would be noticeable.
I'm not remotely cutting edge (just switched from Cursor to Codex CLI, have no fancy tooling infrastructure, am not even vaguely considering git worktrees as a means of working), but Opus 4.5 and 5.2 Codex are both so clearly more competent than previous models that I've started just telling them to do high-level things rather than trying to break things down and give them subtasks.
If people are really set in their ways, maybe they won't try anything beyond what old models can do, and won't notice a difference, but who's had time to get set in their ways with this stuff?
I mostly agree, but today, Opus 4.5 via Claude code did something pretty dumb stuff in my codebase— N queries where one would do, deep array comparison where a reference equality check would suffice, very complex web of nested conditionals which a competent developer would have never written, some edge cases where the backend endpoints didn’t properly verify user permissions before overwriting data, etc.
It’s still hit or miss. The product “worked” when I tested it as a black box, but the code had a lot of rot in it already.
Maybe that stuff no longer matters. Maybe it does. Time will tell.
I don't think they generally one-shot the tasks; but they do them well enough that you can review the diff and make requests for changes and have it succeed in a good outcome more quickly than if you were spoon-feeding it little tasks and checking them as you go (as you used to have to do).
I have had a lot of success lately when working with Opus 4.5 using both the Beads task tracking system and the array of skills under the umbrella of Bad Dave's Robot Army. I don't have a link handy, but you should be able to find it on GitHub. I use the specialized skills for different review tasks (like Architecture Review, Performance Review, Security Review, etc.) on every completed task in addition to my own manual review, and I find that that helps to keep things from getting out of hand.
As someone who’s responsible for some very clean codebases and some codebases that grew over many years, warts and all, I always wonder if being subjected to large amounts of not-exactly-wonderful code has the same effect on an LLM that it arguably also has on human developers (myself included occasionally): that they subconsciously lower their normally high bar for quality a bit, as in „well there‘s quite some smells here, let’s go a bit with the flow and not overdo the quality“.
What's the difference between using llms now vs the first half of 2025 among the best users?
Coding agents and much better models. Claude Code or Codex CLI plus Claude Opus 4.5 or GPT 5.2 Codex.
The latest models and harnesses can crunch on difficult problems for hours at a time and get to working solutions. Nothing could do that back in ~March.
I shared some examples in this comment: https://news.ycombinator.com/item?id=46436885
Ok I will bite.
Every single example you gave is in a hobby project territory. Relatively self-contained, maintainable by 3-4 devs max, within 1k-10k lines of code. I've been successfully using coding agents to create such projects for the past year and it's great, I love it.
However, lots of us here work on codebases that are 100x, 1000x the size of these projects you and Karpathy are talking about. Years of domain specific code. From personal experience, coding agents simply don't work at that scale the same way they do for hobby projects. Over the past year or two, I did not see any significant improvement from any of the newest models.
Building a slightly bigger hobby project is not even close to making these agents work at industrial scale.
That’s right, but it also hints at a solution: split big code bases into parts that are roughly the size of a big hobby project. You’ll need to write some docs to be effective at it, which also helps agents. CICD means continuous integration continuous documentation now.
Most of the stuff I'm talking about here came out in November. There hasn't been much time for professional teams to build new things with it yet, especially given the holidays!
I was going back and looking at timelines, and was shocked to realize that Claude Code and Cursor's default-to-agentic-mode changes both came out in late February. Essentially the entire history of "mainstream" agentic coding is ten months old.
(This helps me understand better the people who are confused/annoyed/dismissive about it, because I remember how dismissive people were about Node, about Docker, about Postgres, about Linux when those things were new too. So many arguments where people would passionately talk about all those things were irredeemably stupid and only suitable for toy/hobby projects.)
> academic papers take 6-12 months to come out
It takes about 6 months to figure out how to get LaTeX to position figures where you want them, and then another 6 months to fight with reviewers
Couldn't AI help with the LaTeX?
Cutting it down to 6 minutes
It feels like we're doing another lift to a higher level of abstraction. Whereas we had "automatic programming" and "high level programming languages" free us from assembly, where higher level abstractions could be represented without the author having to know or care about the assembly (and it took decades for the switch to happen), we now once again get pulled up another layer.
We're in the midst of another abstraction level becoming the working layer - and that's not a small layer jump but a jump to a completely different plane. And I think once again, we'll benefit from getting tools that help us specify the high level concepts we intend, and ways to enforce that the generated code is correct - not necessarily fast or efficient but at least correct - same as compilers do. And this lift is happening on a much more accelerated timeline.
The problem of ensuring correctness of the generated code across all the layers we're now skipping is going to be the crux of how we manage to leverage LLM/agentic coding.
Maybe Cursor is TurboPascal.
The title is doing a lot of work here. What resonated with me is the shift from “writing code” to “steering systems” rather than the hype framing. Senior devs already spend more time constraining, reviewing, and shaping outcomes than typing syntax. AI just makes that explicit. The real skill gap isn’t prompt cleverness, it’s knowing when the agent is confidently wrong and how to fence it in with tests, architecture, and invariants. That part doesn’t scale magically.
Is anyone else getting more mentally exhausted by this? I get more done, but I also miss the relaxing code typing in the middle of the process.
I think there are two groups of people emerging. deep / fast / craft-and-decomposition-loving vs black box / outcome-only.
I've seen people unable to work at average speed on small features suddenly reach above average output through a llm cli and I could sense the pride in them. Which is at odds with my experience of work.. I love to dig down, know a lot, model and find abstractions on my own. There a llm will 1) not understand how my brain work 2) produce something workable but that requires me to stretch mentally.. and most of the time I leave numb. In the last month I've seen many people expressing similar views.
ps: thanks everybody for the answers, interesting to read your pov
I get what you're saying, but I would say that this does not match my own experience. For me, prior to the agentic coding era, the problem was always that I had way more ideas for features, tools, or projects than I had the capacity to build when I had to confront the work of building everything by hand, also dealing with the inevitable difficulties in procrastination and getting started.
I am a very above-average engineer when it comes to speed at completing work well, whether that's typing speed or comprehension speed, and still these tools have felt like giving me a jetpack for my mind. I can get things done in weeks that would have taken me months before, and that opens up space to consider new areas that I wouldn't have even bothered exploring before because I would not have had the time to execute on them well.
The sibling comments (from remich and sanufar) match my experience.
1. I do love getting into the details of code, but I don't mind having an LLM handle boilerplate.
2. There isn't a binary between having an LLM generate all the code and writing it all myself.
3. I still do most of the design work because LLMs often make questionable design decisions.
4. Sometimes I simply want a program to solve a problem (outcome-focused) over a project to work on (craft-focused). Sometimes I need a small program in order to focus on the larger project, and being able to delegate that work has made it more enjoyable.
> I do love getting into the details of code, but I don't mind having an LLM handle boilerplate.
My usual thought is that boilerplate tells me, by existing, where the system is most flawed.
I do like the idea of having a tool that quickly patches the problem while also forcing me to think about its presence.
> There isn't a binary between having an LLM generate all the code and writing it all myself. I still do most of the design work because LLMs often make questionable design decisions.
One workflow that makes sense to me is to have the LLM commit on a branch; fix simple issues instead of trying to make it work (with all the worry of context poisoning); refactor on the same branch; merge; and then repeat for the next feature — starting more or less from scratch except for the agent config (CLAUDE.md etc.). Does that sound about right? Maybe you do something less formal?
> Sometimes I simply want a program to solve a purpose (outcome-focused) over a project to work on (craft-focused). Sometimes I need a small program in order to focus on the larger project, and being able to delegate that work has made it more enjoyable.
Yeah, that sounds about right.
I think for me, the difference really comes down to how much ownership I want to take in regards to the project. If it’s something like a custom kernel that I’m building, the real fun is in reading through docs, learning about systems, and trying to craft the perfect abstractions; but if it’s wiring up a simple pipeline that sends me a text whenever my bus arrives, I’m happy to let an LLM crank that out for me.
I’ve realized that a lot of my coding is on this personal satisfaction vs utility matrix and llms let me focus a lot more energy onto high satisfaction projects
> deep / fast / craft-and-decomposition-loving vs black box / outcome-only
As a (self-reported) craft-and-decomposition lover, I wouldn't call the process "fast".
Certainly it's much faster than if I were trying to take the same approach without the same skills; and certainly I could slow it down with over-engineering. (And "deep" absolutely fits.) But the people I've known that I'd characterize as strongly "outcome-only", were certainly capable of sustaining some pretty high delta-LoC per day.
That's kind of the point here. Once a dev reached a certain level, they often weren't doing much "relaxing code typing" anyways before the AI movement. I don't find it to be much different than being a tech lead, architect, or similar role.
As a former tech lead and now staff engineer, I definitely agree with this. I read a blog post a couple of months ago that theorized that the people that would adopt these technologies the best were people in the exact roles that you describe. I think because we were already used to having to rely on other people to execute on our plans and ideas because they were simply too big to accomplish by ourselves. Now that we have agents to do these things, it's not really all that different - although it is a different management style working around their limitations.
Exactly. I've been a tech lead, have led large, cross-org projects, been an engineering manager, and similar roles. For years, when mentoring upcoming developers what I always to be the most challenging transition was the inflection point between "I deliver most of my value by coding" to "I deliver most of my value by empowering other people to deliver". I think that's what we're seeing here. People who have made this transition are already used to working this way. Both versions have their own quirks and challenges, but at a high level it abstracts.
LLMs are just a programming language/compiler/REPL, though, so there is nothing out of the ordinary for developers. Except what is different is the painfully slow compile time to code ratio. You write code for a few minutes... and then wait. Then spend a few more minutes writing code... and then wait. That is where the exhaustion comes from.
At least in the olden days[1] you could write code for days before compiling, which reduced the pain. Long compilation times has always been awful, but it is less frustrating when you could defer it until the next blue moon. LLMs don't (yet) seem to be able to handle that. If you feed them more than small amounts of code at a time and they quickly go off the rails.
With that said, while you could write large amounts of code and defer it until the next blue moon, it is a skill to be able to do that. Even in C++, juniors seem to like to write a few lines of code and then turn to compiling the results to make sure they are on the right track. I expect that is the group of people who is most feeling at home with LLMs. Spending a few minutes writing code and then waiting on compilation isn't abnormal for them. They don't know any different!
[1] https://xkcd.com/303/
Ya know, I have to admit feeling something like this. Normally, the amount of stuff I put together in a work day offers a sense of completion or even a bit of a dopamine bump because of a "job well done". With this recent work I've been doing, it's instead felt like I've been spending a multiplier more energy communicating intent instead of doing the work myself; that communication seems to be making me more tired than the work itself. Similar?
You’re possibly not entering into the flow state anymore.
Flow is effortless. and it is rejuvenating.
I believe:
While communication can be satisfying, it’s not as rejuvenating as resting in our own Being and simply allowing the action to unfold without mental contraction.
Flow states.
When the right level of challenge and capability align and you become intimate with the problem. The boundaries of me and the problem dissolve and creativity springs forth. Emerging satisfied. Nourished.
It feels like we all signed up to be ICs, but now we’re middle managers and our reports are bots.
> and our reports are bots.
With no gossip, rivalry or backstabbing. Super polite and patient, which is very inspiring.
We also brutally churning them by "laying off" the previously latest model once the new latest is available.
I forget where I saw this (a Medium post, somewhere) but someone summed this up as "I didn't sign up for this just to be a tech priest for the machine god".
Someone commented yesterday that managers and other higher-ups are "already ok with non-deterministic outputs", because that's what engineers give them.
As a manager/tech-lead, I've kind of been a tech priest for some time.
This is why I think LLMs will make us all a LOT smarter. Raw code made it so we stopped heavily thinking in between but now it's just 100% the most intense thought processes all day long.
Yes, absolutely, I can be mentally wiped out by lunch.
I think it's the serial waiting game and inevitable context switching while you wait.
Long iteration cycles are taxing
For me it's the opposite, I'm wasting less energy over debugging silly bugs and fighting/figuring out some annoying config.
But it does feel less fulfilling I suppose.
Nah, I don’t miss at all typing all the tests, CLIs, and APIs I’ve created hundreds of times before. I dunno if I it’s because I do ML stuff, but it’s almost all “think a lot about something, do some math, and and then type thousands of lines of the same stuff around the interesting work.”
I like to alternate focusing on AI wrangling and writing code the old fashioned way.
It's difficult to steer complex systems correctly, because no one has a complete picture of the end goal at the outset. That's why waterfall fails. Writing code agentically means you have to go out of your way to think deeply about what you're building, because it won't be forced on you by the act of writing code. If your requirements are complex, they might actually be a hindrance because you're going have to learn those lessons from failed iterations instead of avoiding them preemptively.
Does using an LLM to craft Hackernews comments count as "steering systems"?
You're totally right! It's not steering systems -- it's cooking, apparently
"it’s knowing when the agent is confidently wrong and how to fence it in with tests, architecture, and invariants."
Strongly suspect this is simply less efficient than doing it yourself if you have enough expertise.
The stereotype that writing code is for junior developers needs to die. Some devs are hired with lofty titles specifically for their programming aptitude and esoteric systems knowlege, not to play implementation telephone with inexperienced devs.
I don't think that anyone actually believes that writing code is only for junior developers. That seems to be a significant exaggeration at the very least. However, it is definitely true that most organizations of this size are hiring people into technical lead, staff engineer, or principal engineer roles are hiring those people not only for their individual expertise, or ability to apply that expertise themselves, but also for their ability to use that expertise as a force multiplier to make other less experienced people better at the craft.
In my world there are Hard Problems that need to be solved for bu$ine$$ rea$on$, no being a "force multiplier" required (whatever that really means).
> Takeaway 3c: Experienced developers disagree about using agents for software planning and design. Some avoided agents out of concern over the importance of design, while others embraced back-and-forth design with an AI.
Im in the back-and-forth camp. I expect a lot of interesting UX to develop here. I built https://github.com/backnotprop/plannotator over the weekend to give me a better way to review & collaborate around plans - all while natively integrated into the coding agent harness.
> Most Recent Task for Survey
> Number of Survey Respondents
> Building apps 53
> Testing 1
I think this sums up everybody complaints about AI generated code. Don't ask me to be the one to review work you didn't even check.
Yea. Nobody wants to be a full-time code reviewer.
Hi it's me, the guy who wants to be a full-time code reviewer.
If you really did that full time and never wrote code, you’d be a terrible reviewer.
Be careful what you wish for.
The new layer of abstraction is tests. Mostly end-to-end and integration tests. It describes the important constraints to the agents, essentially long lived context.
So essentially what this means is a declarative programming system of overall system behavior.
The "Ai-assisted programming" mistaken for "vibe coding" is getting old and annoying
we've never seen a profession drive themselves so aggressively to irrelevance. software engineering will always exist, but it's amazing the pace to which pressure against the profession is rising. 2026 will be a very happy new year indeed for those paying the salaries. :)
We've been giving our work away to each other for free as open source to help improve each other's productivity for 30+ years now and that's only made our profession more valuable.
I see little proof open source has resulted in higher wages and not the fact that everything is being digitized and the subsequent demand for such people to assist in such.
I'm not sure how I can prove it, but ~25 years ago building software without open source sucked. You had to build everything from scratch! It took months to get even the most basic things up and running.
I think open source is the single most important productivity boost to our industry that's ever existed. Automated testing is a close second.
Google, Facebook, many others would not have existed without open source to build on.
And those giants and others like them that were enabled by open source employed a TON of people, at competitive rates that greatly increased our salaries.
25 years ago, I was slinging apps together super fast using VB6. It was awesome. It was a level of productivity few modern stacks can approach.
Agentic coding is just another rhyme of 25 y/o frenzy of "let's outsource everything to India." The new generation thinks this time is really special with us. Let's check again in 25 years
> 25 years ago, I was slinging apps together super fast using VB6. It was awesome. It was a level of productivity few modern stacks can approach.
If that were too, wouldn't we all be using VB today?
Being productive with VB6 (final release in 1998) is very far from being productive with modern stacks. Apples and orangutans.
Will I grant advantages to certain aspects of VB6? Sure. Could some lessons be applicable today? Probably. But just like historians say, don't make the mistake of ignoring context when you compare things from different eras.
P.S. Search for "most dreaded legacy languages" and you'll see a lot of discussion on VB6, much of it unflattering. I don't have a dog in that fight -- I can't tell if it is more on polarizing end of things (some love, some hate) or more on the "universally loathed" side.
I'm too young to have used VB in the workforce, but I did use it in school, and honestly off that alone I'm inclined to agree.
I've seen VB namedropped frequently, but I feel like I've yet to see a proper discussion of why it seems like nothing can match its productivity and ease of use for simple desktop apps. Like, what even is the modern approach for a simple GUI program? Is Electron really the best we can do?
MS Access is another retro classic of sorts that, despite having a lot of flaws, it seems like nothing has risen to fill its niche other than SaaS webapps like airtable.
You can add Macromedia Flash to that list - nothing has really replaced it, and as a result the world no longer has an approachable tool for building interactive animations.
Indeed it did; I remember those times. All else being equal I still think SWE salaries on average would of been higher if we kept it like that given basic economics - there would of been a lot less people capable of doing it but the high ROI automation opportunities would of still been there. The fact that "it sucked" usually creates more scarcity on the supply side; which all being equal means higher wages and in our capitalist society - status. Other professions that are older as to the parent comment already know this and don't see SWE as very "street smart" disrupting themselves. I've seen articles recently like "at least we aren't in coding" from law, accounting, etc an an anecdote to this.
With AI at least locally I'm seeing the opposite now - less hiring, less wage pressure and in social circles a lot less status when I mention I'm a SWE (almost sympathy for my lot vs respect only 5 years ago). While I don't care for the status aspect, although I do care for my ability to earn money, some do.
At least locally inflation adjusted in my city SWE wages bought more and were higher in general compared to others in the 90's-2000's than on wards (ex big tech). Partly because this difficulty and low level knowledge meant only very skilled people could participate.
> ex big tech
I mean, this seems like a pretty big thing to leave out, no? That's where all the crazy high salaries were!
Also, there are still legacy places that more or less build software like it's 1999. I get the impression that embedded, automotive, and such still rely a lot on proprietary tools, finicky manual processes, low level languages (obviously), etc. But those are notorious for being annoying and not very well paid.
I'm talking about what I perceive to be the median salary/conditions with big tech being only a part of that. My point is more that I remember back in that period good salaries could be had outside big tech too even in the boring standard companies that you state. I remember banks, insurance, etc paying very well for example compared to today for an SWE/tech worker - the good opportunities seemed more distributed. For example I've seen contract rates for some of the people we hire haven't really changed for 10 years for developers. Now at best they are on par with other professional white collar workers; and the competition seems fiercer (e.g. 5 interviews for a similar salary with leetcode games rather than experienced based interviews).
Making software easier and more abstract has allowed less technical people into the profession, allowed easier outsourcing, meant more competition/interview prep to filter out people (even if the skills are not used in the job at all), more material for AI to train on, etc. To the parent comment's point I don't think it has boosted salaries and/or conditions on average for the SWE - in the long run (10 years +) it could be argued that economically the opposite has occurred.
Monopolizing the work doesn't work unless you have the power to suppress anyone else joining the competition, i.e. "certified developers only".
Otherwise people would have realized they can charge 3x as much by being 5x as productive with better tools while you're writing your code in notepad for maximum ROI, and you would have either adjusted or gone out of business.
Increased productivity isn't a choice, it's a result of competition. And that's a good thing overall, even if it sucks for some developers who now have to actually work for the first time in decades. But it's good for society at large, because more things can be done.
Sure - I agree with that, and I agree its good for society but as you state probably not as good for the SWE who has to work harder for the same which was my point and I think you agree. Other professions have done what you have stated (i.e. certification) and seen higher wages than otherwise which also proves my point. They see this as the "street smart" thing to do, and generally society respects them for it putting their profession on a higher pedestal as a result. People respect people who take care of themselves first generally I find as well. Personally I think there should be a balance between the two (i.e. a fair go for all parties; a fair day's work with some job security over a standard career lifetime but not extortionary).
Also your notion of "better tools" may of not happened, or happened more slowly without open source, AI, etc which would of meant higher salaries for longer most probably. That's where I disagree with the parent poster's claim of higher salaries - AI seems to be a great recent example of "better tools" disrupting the premium SWE's enjoy rather than improving their salaries. Whether that's fair or not is a different debate.
I was just doubting the notion of the parent comment that "open source software" and "automated testing" create higher salaries. Usually efficiency economically (some exceptional cases) creates lower salaries for the people who are made more efficient all else being equal - and the value shifts from them to either consumers or employers.
even if that's true it's clear enough AI will reduce the demand for swe
I don't think that's certain. I'm hoping for a Jevons paradox situation where AI drives down the cost of producing software to the point that companies that previously weren't in the market for custom software start hiring software engineers. I think we could see demand go up.
This makes sense. Imagine PHP or NodeJS without a framework, or front end development without React. Your projects would take much longer to build. The time saved with the open source frameworks and libraries is more than what an AI agent can save you.
> we've never seen a profession drive themselves so aggressively to irrelevance.
Should we be trying to put the genie back in the bottle? If not, what exactly are you suggesting?
Even if we all agreed to stop using AI tools today, what about the rest of world? Will everybody agree to stop using it? Do you think that is even a remote possibility?
Does the rest of the world want to make money in a way not involving digging ditches? I feel like people from developing countries that spend 18 hours a day studying, giving their entire childhood to some standardized test, may not want yo be rewarded with no job prospects. Maybe that’s a crazy position.
Don't care have too much to do must automate away my today responsibilities so I can do more tomorrow trvst the plqn
Also it really baffles me how many are actually in on the hype train. Its a lot more than the crypto bros back in the day. Good thing AI still cant reason and innovate stuff. Also leaking credentials is a felony in my country so I also wont ever attach it to my codebases.
I think the issue is folks talk past each other. People who find coding agents useful or enjoyable are labeled “on the hype train” and folks for which coding agents don’t work for them or their workflow are considered luddites. There are an incredible number of contradicting claims and predictions out there as well, and I believe what we see is folks projecting their reaction to some amalgamation of them onto others. I see a lot of “they” language, and a lot of viral articles about business leadership “shoving AI down our throats” and it becomes a divisive issue like American political scene with really no one having a real conversation
I think the reason for the varying claims and predictions is because developers have wildly different standards for what constitutes working code. For the developers with a lower threshold, AI is like crack to them because gen ai's output is similar to what they would produce, and it really is a 10x speedup. For others, especially those who have to fix and maintain that code, it's more like a 10x slowdown.
Hence why you have in the same thread, some developer who claims that Claude writes 99% of their code and another developer who finds it totally useless. And of course others who are somewhere in the middle.
Have you considered that it's a bit dismissive to assume that developers who find use out of AI tools necessarily approve of worse code than you do, or have lower standards?
It's fine to be a skeptic. Or to have tried out these tools and found that they do not work well for your particular use case at this moment in time. But you shouldn't assume that people who do get value out of them are not as good at the job as you are, or are dumber than you are, or slower than you are. That's just not a good practice and is also rude.
I never said anything about being worse, dumber, and definitely not slower. And keep in mind worse is subjective - if something doesn't require edge case handling or correctness, bugs can be tolerated etc, then something with those properties isn't worse is it?
I'm just saying that since there is such a wide range of experiences with the same tools, it's probably likely that developers vary on their evaluations of the output.
Okay, I certainly agree with you that different use cases can dictate different outcomes when using AI tooling. I would just encourage everyone who thinks similar to you to be cautious about assuming that someone who experiences a different result with these tools is less skilled or dealing with a less difficult use case - like one that has no edge cases or has greater tolerance for bugs. It's possible that this is the case, but it is just as possible that they have found a way to work with these tools that produces excellent output.
Yeah I agree, it doesn't really have to do with skill or different use cases, it's just what your threshold is for "working" or "good".
There's also the effect of different models. Until the most recent models, especially for concise algorithms, I felt it was still easier to sometimes do it myself (i.e. a good algo can be concise/more concise than a lossy prompt) and leave the "expansion/repetitive" boilerplate code to the LLM. At least for me the latest models do feel like a "step change" in that the problems can be bigger and/or require less supervision on each problem depending on the tradeoff you want.
Its all a hype train though. People still believe in the AI gonna bring utopia bullshit while the current infra is being built on debt. The only reason it still exists is that all these AI companies believe in some kind of revenue outside of subscriptions. So its all about:
Owning the infrastructure and enshittify (ads) once enough products are based on AI.
Its the same chokehold Amazon has on its Vendors.
your credentials shouldn't be in your codebase to begin with!
.env files are a thing in tons of codebases
If your secrets are in your repo, you've probably already leaked them.
but thats at runtime, secrets are going to be deployed in a secure manner after the code is released
.env files are used to develop as well, for some things like PayPal u dont have to change the credentials, you just enable sandbox mode. If I had some LLM attached to my codebase, it would be able to read those credentials from the .env file.
This has nothing to do with deployment. I never talked about deployment.
If you have your PayPal creds in your repository, you are doing it wrong.
Is the title an ironic play on AI’s trademark writing style, is it AI generated, or is the style just rubbing off on people?
I think it’s a popular style before gen ai and the training process of LLMs picked up on that.
That’s not how LLMs work, it’s part of the reinforcement learning or SFT dataset, data labelers would have written or generated tons of examples using this and other patterns (all the emoji READMEs for example) that the models emulate. The early ones had very formulaic essay style outputs that always ended with “in conclusion”, lots of the same kind of bullet lists, and a love of adjectives and delving, all of which were intentionally trained in. It’s more subtle now but it’s still there.
Maybe I was being imprecise, but I’m not sure what you mean by “not how LLMs work” - discovering patterns of how humans write is exactly the signal they are trained against. Either explicitly curated like SFT or coaxed out during RLHF, no?
It could even have been picked up in pretraining and then rewarded during rlhf when the output domain was being refined; I haven’t used enough LLMs before post training to know what step it usually becomes noticeable.
Excellent survey, but one has to be careful when participating in such surveys:
"I’m on disability, but agents let me code again and be more productive than ever (in a 25+ year career). - S22"
Once Social Security Administration learns this, there goes the disability benefit...
I think you eventually lose disability benefits anyway once you start making money.
I often tell people that agentic programming tools are the best thing since cscope. The last 6 months I have not used cscope even once after decades of using it nearly daily.
[0] https://en.wikipedia.org/wiki/Cscope
You know what. After seeing all these articles about AI/LLM for these past 4 years, about how they are going to replace me as software developers and about how I am not productive enough without using 5 agents and being a project manager.
I. Don't. Care.
I don't even care about those debates outside. Debates about do LLM work and replace programmers? Say they do, ok so what?
I simply have too much fun programming. I am just a mere fullstack business line programmer, generic random replaceable dude, you can find me dime a dozen.
I do use LLM as Stack Overflow/docs replacement, but I always code by hand all my code.
If you want to replace me, replace me. I'll go to companies that need me. If there are no companies that need my skill, fine, then I'll just do this as a hobby, and probably flip burgers outside to make a living.
I don't care about your LLM, I don't care about your agent, I probably don't even care about the job prospects for that matter if I have to be forced to use tools that I don't like and to use workflows I don't like. You can go ahead find others who are willing to do it for you.
As for me, I simply have too much fun programming. Now if you excuse me, I need to go have fun.
I simply will not spend my life begging and coaxing a machine to output working code. If that is what becomes of this profession, I will just do something else :)
If I wanted to do that, I'd just move into engineering management and work with something less temperamental and predictable - humans.
I'd at least be more likely to get a boost in impact and ability to affect decision making, maybe.
Until you realize you're just begging and coaxing a human to better beg and coax a machine to output working code - when you could just beg and coax the machine yourself.
At least I'd be the one interfacing with a human instead of a machine :P
It would definitely be the profession if we stopped developing things today. Think about the idea of coding agents 2 years ago, I personally found them very unrealistic and am now coding exclusively with them despite them being either a neutral or net negative to my development time simply because I see the writing on the wall that in 6 mos to a year they will probably be a huge net positive and in 2-3 years the dismissive attitude towards adoption will start to look kind of silly (no offense). To me we are _just_ at the inflection point where using and not using coding agents are both totally sensible decisions.
Hear hear. I didn't spend half my life getting an education, competing in the corporate crab bucket, retraining and upskilling just to turn into a robot babysitter.
I hear you but I feel like you (and really others like you, in mass) should not be so passive about your replacement. For most programmers, simply flipping burgers for money to enjoy programming a few hours a week is not going to work. Making a living is a thing. If you are reduced to having to flip burgers that means the economy will gave collapsed and there won’t be any magic Elon UBI money to save us.
Easy to say if you either:
(1) already have enough money to survive without working, or
(2) don't realize how hard of a life it would be to "flip burgers" to make a living in 2026.
We live very good lives as software developers. Don't be a fool and think you could just "flip burgers" and be fine.
Ah, I actually did flip burgers. So I know.
I also did dry cleaning, cleaning service, deli, delivery guy, etc.
Yup I now have enough money to survive without working.
But I also am very low maintenance, thanks to my early life being raised in harsh conditions.
I am not scared to go back flipping burgers again.
"Yup I now have enough money to survive without working" Your opinion is borderline irrelevant then.
having fun isn't tied to employment unless you are self-employed even then what's fun should not be the driving force
"get a job doing something you enjoy and you'll never work a day in your life"
or something like that
That sounds miserable to me :(
you work on somebody's dime, its no longer your choice
it's your choice whose dime you work on. they can compete for your work by making it fun for you.
sure unemployment is also a choice
fun work > tedious work > unemployment
not sure why so many people feel like factoring fun into what job you want to take is so unthinkable, or that it's just a false dichotomy between the ideal job and unemployment
you are describing the ideal which is not a reality for many many people as it is not common
it's a trade-off; you need a job but you typically interview at several places, collect offers, and weigh them according to various criteria. all the pro-fun posters are saying is that "enjoy the job" is a very highly ranked criterion for us.
It's my life, it's my choice.
Why? It is a matter of values. Fun can be a driving force just like money and stability is. It is simply a matter of your values (and your sacrifices).
Like I said, I am just a generic replaceable dime a dozen programmer dude.
you dont get paid to have fun but to produce as a laborer
a job isn't supposed to be fun its nice when it is but it shouldn't be what drives decisions
You mean it shouldn't be the driving force of your employer to make decision. Yes I agree 10000%
I meant it can be your (not necessarily your employer) driving decision in life.
Of course, you need to suffer. That's about having tradeoffs.
almost all employers are going to expect you to use AI and produce more with it
you can definitely choose not to participate and give the opportunity someone who are happy to use AI and still have fun with it.
most organizations have awful leadership, sure
but that doesn't mean you can't (or shouldn't) work around it
have you tried telling your boss you won't use the AI anymore while the rest of the team uses it ?
how do you imagine such conversation to play out im curious
> Through field observations (N=13) and qualitative surveys (N=99)...
Not a statistically significant sample size.
97 samples is enough to get a 95% confidence level if you accept a 10% margin of error. 99 is not so bad, at least.
https://www.surveymonkey.com/mp/sample-size-calculator/
This is a qualitative methods paper, so statistical significance is not relevant. The rough qualitative equivalent would instead be "data saturation" (responses generally look like ones you've received already) and "thematic saturation" (you've likely found all the themes you will find through this method of data collection). There's an intuitive quality to determining the number of responses needed based on the topic and research questions, but this looks to me like they have achieved sufficient thematic saturation based on the results.
Significance depends on effect size.
Same thoughts exactly.
I haven't seen the definition of an agent, in the paper. Do they differentiate agents from generic online chat interfaces?
Page 2: We define agentic tools or agents as AI tools integrated into an IDE or a terminal that can manipulate the code directly (i.e., excluding web-based chat interfaces)
An agent takes actions. Chat bots only return text.
I like to think of it as "maintaining fertile soil"
Don’t let anyone tell you the right way to program a computer.
Do it in the way that makes you feel happy, or conforms to organizational standards.
The right way to program a computer:
Well
Idk, I still mostly avoid using it and if I do, I just copy and paste shit into the Claude web version. I wont ever manage agents as that sounds just as complicated as coding shit myself.
It's not complicated at all. You don't "manage agents". You just type your prompt into an terminal application that can update files, read your docs and run your tests.
As with every new tech there's a hell of a lot of noise (plugins, skills, hooks, MCP, LSP - to quote Kaparthy) but most of it can just be disregarded. No one is "behind" - it's all very easy to use.