I feel his pain. I am more towards the opposite end of the spectrum [1]. I program to get things done. Usually, I don’t like programming. It’s too focused on one thing. Sometimes I like it though, precisely because it’s focused on one thing. But I love the things you can imagine and build with it. So for me, LLMs are amazing because I get to be an idea guy when I want but sometimes I can go deep, also when I want. I’ve kept up too, so I have the experience backing it up.
With that said, when I read this blog post. I feel the author’s pain. And it’s the first time that I emotionally get what the other side of the programming spectrum feels what it has lost. I feel sad about it. And because of it, I will also wonder about ways of bringing it back.
[1] not fully though. Something I can love coding by hand for months.
> So for me, LLMs are amazing because I get to be an idea guy
Enjoy this while it lasts, because very shortly LLMs are going to be fundamentally better at being the "idea guy" as well.
I had the same feeling reading this article today as I did reading the NYT article about the closing of Butter Ridge dairy farm. All that family wants to do is raise cows, name them, and milk them but it doesn't work as a business anymore because the factory farm machine took the viability away.
I feel the author deeply; this is coming for everyone across most of the economy.
> Enjoy this while it lasts, because very shortly LLMs are going to be fundamentally better at being the "idea guy" as well.
Maybe for the ideas that are so far from novel that there's a corpus of training data that could train an LLM to reproduce it. But for actually novel ideas, LLMs won't ever be much use. They can't interact with the world directly, they can only interact with text (or I guess bits).
And I guess that's not much of a distinction, as truly novel good ideas are very rare and you can go very far applying a good idea into a new domain. But at the edges where true novelty is required, LLMs will either hallucinate or guide you away from the edge.
So much comes down to timing - lucky author that got to experience it for 30 whole years whereas some of us had a scant few to enjoy building as it was, flow state, and some semblance of security. At least I made plans the last 5 years in anticipation of change, can't imagine the sheer dislocation of just entering the job market now.
Pros and cons. Some of the people who were lucky to enjoy those 30 years are also emotionally being hit the hardest right now, and if life threw a few curveballs at you along the way you don't necesarily have attained the sort of stability where you don't have to worry, either. Plus ageism can make it even harder to pivot.
I have programmer friends in their 40s to 60s who are seriously depressed currently (and 20 year olds worried for theirnl future perspectives, of course). Mental health is not just a young person's game.
I strangely feel quite lucky that I got more and more into electronics and hardware over time as I moved from web and desktop more and more into embedded/consumer electronics and companies who also employ mechanical and EE engineers. When I was younger I used to dumbly worry this meant giving something up (the purity of software approaches, etc.), but instead it made me consider myself an Engineer with a capital E and strive to learn the engineering method more generally, and learn so many other fields of the trade. It turns out this is a much more resilient identity than just Programmer and I recommend that approach.
It's funny that you mention EE. When I went back to school for CS in the early 2000s (I was a philosophy major... dumb), I was taking night classes with EEs who were fleeing to software. This was in the Boston area. A large contingent of my classmates worked at Avid. They started moving jobs out of the US and employees were told that if they switched to software they could remain employed. The mood was pretty grim, and these folks commented to me more than once that I was lucky that I loved writing software, because largely, they didn't.
Right, I also hear that sentiment pretty often. I don't regret having a software career and am very glad for my 25 years of C++ and many other things. I wouldn't want to be without that, and it probably did and does pay better.
However, it's pretty nice that these days I can also swing a semi-decent PCB, know my way around scopes and logic analyzers quite well, CAD something up for DFM in a number of processes from thermoplastics to machining, taught myself a fair bit of structural engineering, set up a FEM analysis correctly, etc. If only because it lets me bridge worlds and tie software and hardware together more effectively in the projects I'm in.
I cannot do any of these things as well as a seasoned veteran, but it has given me a broader appreciation of engineering overall and the commonalities between it all, to the point where I can also muster up leadership in engineering orgs more broadly and am not as hurt over the prospect that my pure programming skills might get devalued or diluted, or change.
For example, software engineers generally scoff at the perceived crustyness and lack of agility in classical mechanical engineering processes, but on the other hand mechanical engineering is far more experienced at defense-in-depth type approaches, dealing in components that have a failure rate to them and designing with error bars and safety factors, and I find some of that mindset has transferred quite naturally to engineering with our unreliable LLM friends at scale the past two years.
It takes a lot of the sting out if listening to Phish isn't your only move. Well, maybe not a lot, but at least it doesn't get so existential. Don't be a Programmer, be an Engineer. It's a lot easier to feel useful during a time of much doubt.
Everyone is talking about the agents part. I'm going to praise this post for describing it very clearly that some people, from a young age, don't need phases, growing up, trying things, figuring out, exploring the world, finding themselves.
Some people are just born something (engineers, in this case), and they're that something for life.
I always have a hard time explaining to "normal people" that such life is not boring at all, in fact, I can't remember a single time in my life where I was actually "bored".
The people who go through phases are also born to be something for life: adaptable, and used to change. Being good at doing exactly one thing your entire life a certain way has the potentially fatal flaw of having significant issue doing anything else.
I feel like I've lived 3-4 completely different lives so far, but the constant is the ability to adapt to the next one, and still find joy in it while you're there. "Survival is the ability to swim in strange water."
Personally, the AI tools have been transformative for work, but haven't affected how I work much. I have always coded as a team. I'd often do the largest and most complicated parts myself, but work (both at work work and my hobby work) has always been about passing things between colleagues based on what our strengths were and how much time we had.
The AI tools are another colleague. They work incredibly fast, and I do less coding myself now, but my goal was always to solve the problem, not the code itself. The AI tools do a great job most of the time, but they sometimes screw up and need more guidance or me to step in to fix the thing (usually a very small error compared to the whole). If they screw it all up, we might need to start from scratch, or I might need to just do it myself. But that's not most of the time. And then I figure out the thing missing in my process to move them in the right direction, and improve it.
I feel like any software developer used to collaborating with, training, and mentoring other developers knows exactly how to work with AI tools, and the only main difference is how much effort I put into being really nice about it the whole time.
One main thing has changed. Before I would handle the most complicated problems people were having trouble with myself. Now I allow AI to work on them, even if I know I'll have to fix it. The difference is that I care about the time, the strain, and the morale of my coworkers. AI is just getting paid for iterating tokens, so I don't have to feel bad about what I ask it to figure out for me.
> Some people are just born something (engineers, in this case), and they're that something for life.
Yes, and it's very not fun when your identity is being reshaped before your eyes in the matter of a couple years.
I wonder how many developers are going through real grief right now, while everybody else, lacking empathy, are just repeating "get a grip, it's just a tool" or "you better adapt or you're done".
Well, I know I have gone through these difficult emotions, and I choose now not to identify with my work, or at least my career any more. I certainly do not identify as what most people these days refer to as 'software engineer' any more.
This is a great reason not to identify too much with your work. I have enjoyed AI because it has reminded me that my real calling is art, and that I should be doing that at 8 pm, not coding
I don’t think anyone’s true calling is coding. That’s like saying you really like the act of writing, so much that you’d become a stenographer or a typist or something where you do zero higher level thinking and just absent mindedly press buttons.
Most people who are good at tech hate coding so much that they come up with elaborate abstractions so that they can avoid doing more of it.
I recommend you read the post because that's a really bad misunderstanding of the mindset, and like the comment at the top of this chain says, the post explains it well.
I read the post. I don’t agree with the “people are born to do one thing” mindset. There’s a lot of possibilities out there for everyone. I do identify with this OP fellow somewhat, except that I usually don’t code for fun on nights and weekends (also sunday code sesh can be fun)
Not sure if your 'enjoyed AI' is meant literally. I have escaped my existential crisis and found solace in art as well, simply because it's a very human method of self-expression and taking shortcuts to pain, effort and creativity by prompting LLMs is still frowned upon, at least by those that take art seriously.
The only way I can still enjoy programming now is if it's applied to artistic endeavours. I'm done with the soulless, cost-efficient software "engineering" (which by itself is a laughable proposition and a far cry from the high standards of other fields of engineering)
I do literally enjoy working with AI sometimes, other times it is hell. But sometimes coding by hand is hell, too, usually when working on other people’s extremely hacky and procedural code.
What I really enjoy is learning. I came back to computing after many years in another field, and was completely in love. Everything finally clicked, I'd spend hours reading everything I could, coding, trying things out, letting half-half-built projects pile up as I discovered new things.
AI has completely ruined this for me. Its boring having someone else do stuff for you. And worse, I feel I'm un-learning at a rapid rate.
The magic has gone, I'm not sure I want to be in this game in another five years.
IMO assisted coding (auto complete style) has more flow state than the old days of getting stuck on obscure bugs (as satisfying as those were to crack).
Full agent coding however is the complete opposite, you’re in constant damage control of a junior who moves fast and breaks everything. They’ve got better but still do dumb mistakes.
lot of engineers are discovering firsthand what it’s like to manage a team of eager but useless employees. Not fun at all
The other thing is as far as I can tell, the power-tool style modes where you use AI to boost focus and do quick research is both much cheaper and, by the time you account for all the damage control/prompt tuning/other externalities, roughly as fast as full agentic.
I suspect with the prices going up, that realization is going to be pretty appealing.
Prices are not going up. DeepSeek V4 Pro is 5-10x cheaper than Claude 4.7.
As some of us have been predicting, model capability had already mostly plateaued, and the Chinese have and will continue to relentlessly push cost down. Chinese models will be used for 95% of things, with nation-native models for security/sovereignity-sensitive workloads. Eventually (5+ years from now), efficiency gains and hardware progress will make running local models the dominant way of doing things.
And yes, that puts the investors of Claude and OpenAI in quite a pickle.
I want the frontier on prem to be true but IMO not good enough yet unless async.
What started as all-you-can-eat $50 buffet has quietly become a $6k bill, frontier models that don’t ship your codebase to Beijing don’t come cheap anymore.
Yes, that’s basically what I’m talking about. Without subsidies, that’s a lot of incentive to use more efficient/open weight models, but also to use them in ways that are less compute intensive—fewer tokens, shorter reasoning chains, less nonstop, and generally tie up less hardware for a lot less time.
I agree, I also think there are long term costs to agentic coding that we won't see yet for a while, gradually and then suddenly.
It feels like nobody is even trying to iterate on the "power-tool style" usage of language models, everyone jumped straight to agents. It's not clear to me that removing a human from the loop is strictly more efficient though. Imagine an editor with an embedded language model (either running locally or using cheap cloud models) that is constantly churning, analysing code, reading debug logs, offering suggestions. Refactoring is not b̴e̴g̴g̴i̴n̴g̴ asking a chat interface to make changes, but structured refactors utilising the language models powered by the underlying ast representations to pull off much larger or involved refactors that surpass the abilities of current IDEs. Or doing codegen in your editor in a more structured (and thus repeatable) way compared to an agent spitting out code that you have to then review.
> IMO assisted coding (auto complete style) has more flow
I find this impossible to wrap my head around. Any time I autocomplete more than a single line, my flow is destroyed when I have to review what was just autocompleted. The only way I could remain in "flow" in this circumstance would be if I didn't review the code... I'm not ready to trust the LLMs that much yet
> IMO assisted coding (auto complete style) has more flow state than the old days of getting stuck on obscure bugs (as satisfying as those were to crack).
I strongly doubt that. Flow state happens when you can disassociate thinking and doing. Most of your brain is on the thinking while the doing lags behind. It’s like driving on a well known road. You don’t pay that much attention to the act of driving. Instead maybe you start planning what you want to do on your next stop
Using ai auto complete would be like your well known road, but turned into an obstacle course. You need your full focus on the road. Not that different from someone that is unfamiliar with the road, but not a flow state either. Full agentic coding is you being in the passenger seat with a driver known to be sometimes erratic. Less thinking but more worrying.
People are blown away when I tell them that, in the last 6 months, my job of coding has changed entirely, and that I now write very little code, but instead manage agents who write it. It is still engineering, and I still very much care what that code is, it's interfaces, how it interacts with the world, how it is tested, etc. etc., but it has taken me a while to get used to the idea of me not writing the code. I'm still not sure how I feel about it, although I am getting more done, and it has helped me keep better focus on "the big picture". That is tough to do when your day is spent in the weeds.
It isn't about the code (they probably don't even know what that is), they focus on the fact that my daily job tasks have changed entirely (iow, I used to do 'something', 'what' doesn't matter, and now I dictate to an AI to do it). Most people can't fathom this unless they change to a completely different profession/job.
I don't code (well, I write scripts for my own personal use and use Emacs), but I follow with interest these reports from software developers. Why? Because my profession is similar screen- and keyboard-based intellectual work, and what has come for devs will probably come for many other careers.
It's worth mentioning that one reason LLMs can clean up on coding tasks is because of the volumes of available data. Not only has the world produced copious volumes of code, they continue to produce copious volumes of code, and some code can even be generated synthetically (ie, not from an LLM, albeit at high cost).
Other domains are not like this. There will probably never be enough poetry out there to make an LLM do anything but be a poor imitation of a poet. This data is extremely hard to generate.
Urm, isn't that a terrible example? LLMs are better at poetry and rhyming then most people trying their best...
Also you're thinking with an extremely short time horizon there.
All jobs which are centered around computers will be impacted the same way programming is today in the medium term. It's just a question of time until the data is gathered and tooling is adjusted. Because every company that's currently employing people to do something with software will start to use that data as training material, and a few years later they'll be swallowed along.
So that means eg bookkeeping is still safe... I sincerely doubt it'll be like that in 2035. It'll probably still be somewhat safe in 2030 though
I suspect most industries aren't quiet filled with people like the author though, most people treat their job as... Well, a job.
This is a wild take. Of all things, you think a large language model will be stumped by poetry?
I would wager everything I had that an AI could already win a blind poetry contest, especially if any effort was put into fine-tuning it. Unbelievable effort has been put into optimizing them for coding given the financials; next to none for poetry.
Reading this, I felt a familiar kind of sadness.
I have also felt some version of this recently: the sense of loss, and the question of whether I am still a “real programmer” if I am no longer writing code in the same way. There is a strange grief in letting go of a skill that once gave you pride.
But when I think about it from the author’s position, I may actually have been lucky.
For this person, writing code may have been a way of life. In my case, I only started doing field work and using AI relatively recently, so I was able to adapt faster than I expected.
If your whole way of life changes, the shock must be much greater.
In contrast, I had no real status or social position to protect, so perhaps it was easier for me to let go. If I had tried to compete fairly and directly, I could not have beaten the experience and accumulated skill of veteran programmers.
Of course, my ability to write code was something I was somewhat proud of. Giving it up was painful, and it brought regret and a sense of inferiority. But at the same time, I also find myself thinking: “Was I really supposed to fight against veterans like this?”
Recently, this feels very similar to Durkheim’s concept of anomie. While reading this, I kept thinking about categories such as conformity, innovation, ritualism, retreatism, and rebellion. There are many points here that make me think.
If, in the future, coding changes again from today’s agent-based coding into some other form, what will happen to me then? By observing how senior programmers are reacting now, perhaps I can draw my own conclusions and prepare for that moment.
Right now, agent-based coding that depends on specific companies is dominant. But I think the current price of agentic coding is too cheap. At some point, when it becomes more expensive, local LLMs may become mainstream. If that happens, damaged or weakened code-writing ability may become necessary again.
So the question is: how should I prepare for that?
I understand that commercial companies want to get as much value out of each developer as possible, I understand that managers want work to complete as quickly as possible. I can see why they are so excited about LLM tooling and the current increase in output.
This post makes a good point: managing LLM models isn't really the same thing as thinking hard about a particular problem, solving that problem and then concretizing it in code. If the work becomes managing models, I think we're going to see a pretty stark divergence between what people enjoy about developing software today and what the job is requiring. I'm not yet convinced LLM tooling will stick but, if it does, it makes me wonder what kind of person will be doing software development. Maybe some of same people and they find something else to enjoy about the job but I bet a lot of a different kind of person.
Personally, I am very uncomfortable with the idea that all software development might be mediated by LLM tooling and, as a consequence, require payment to a large corporate entity like Anthropic or Google. Hopefully some open source projects will remain open to accepting PRs from people, like the author of the OP, who enjoy working in that "flow" state. I enjoy writing software as a hobby (as well as at my job) but it looks like the hobby might become a larger source of personal fulfillment.
> I'm not yet convinced LLM tooling will stick but, if it does, it makes me wonder what kind of person will be doing software development. Maybe some of same people and they find something else to enjoy about the job but I bet a lot of a different kind of person.
TBH, I've often felt like a weirdo who enjoys "the wrong things" about software engineering.
Ditto. I programmed for 30 years with Underworld on. Listening to the beginning of Juanita/Kiteless instantly drops me into flow state. And it no longer fits if I am using an agent. I appear to be almost the same age as the writer of the post; just a little older. And I got that grocery store job.
So I am working on a personal OpenBSD-related project. I have no idea if it will ever see the light of day. It scratches an itch and I am having fun doing it, and I don’t use AI. It’s forcing me to read man pages carefully and OpenBSD source code (side note: the lack of comments in their code is almost a “fuck you”).
I have a similar tendency. I doesn't really work for me to write code and listen to lyrics so EDM and grindcore, like carcass, works for me. Either there are few words or you can understand them.
Programming with Claude is still engineering. It is like designing a bridge, which remains engineering even when a worker pours the concrete instead of you.
In the past we were forced to pour the concrete ourselves. I understand how many of us enjoyed the sound and the smell of the concrete being poured. Myself, I’m happy to never get my hands dirty again, and focus on the actual engineering.
lmao, I think I read it here and then I proceeded to think I came up with it myself. I’m a fraud. (I put “concrete” into the search field and found it)
I was with you until the last three words. Craftsmanship in writing code is also “actual engineering”, it’s just not the engineering that people will pay a human to do now.
Chemical engineer here - I look at this similar to how a lot of professional engineers work, drafting gets passed off to eng techs, pfds get passed off to eng techs... The "actual engineering" is in the ideas, the applications, the reviews, and most importantly the final accountability.
Studied Electronic Engineering and have friends in Civil Engineering. By the time, you get to the actual implementation, the design has been tested so many times that errors are often material defects, not design defects. Engineering is about ironing out all the design defects and have all relevant unknowns be known.
Working with claude is closer to “grandma recipes” than any engineering or scientific practice.
Thanks for this perspective. Do we really miss typing ascii characters into an editor? That seems to me the least consequential and least interesting part of building software systems. Always has been.
Dare I say those stuck on nostalgia for pressing keys are demonstrating that they cared more about their own personal experience than about the outcome of their work? Now that coding is automated, we have to elevate our ambitions.
Ironically, Phish's music emerges from egoless expression (to paraphrase keyboardist Page McConnell). Giving up your own personal stake in the process is literally what brings something as beautiful as Phish's music into existence. We need to do the same with our software; give up the notion that "our" code is meaningful.
Dare I say those stuck on nostalgia for pressing keys are demonstrating that they cared more about their own personal experience than about the outcome of their work? Now that coding is automated, we have to elevate our ambitions.
YES. The beauty of programming is and always was that, first, you enjoyed it and, second, for some oddball reason you could actually get paid to do it. And one can't produce anything good unless you actually love working on it which means you want to put yourself working on it. The outcome might accidentally serve the one who pays for it but ultimately what did get the work finished was the sensation when you were reaching the point where you would finally tie things together and see everything you designed come to life and work together.
AI doesn't give you that personal involvement. We can do it but it's a different line of work and we care very little about what goes in and what comes out. We just do the grunt work of connecting the two ends. We're not for a fuck interested in elevating ambitions which is a word that relates to what is outside while all the good stuff comes from the inside.
I think concentrating on the physical act of typing on the keyboard is maybe taking it a little too far. The author of the OP talks more about holding a lot of the problem in their head and entering a "flow" state where they figure out a solution.
Most of my interaction with AI models and agents is still mediated by a keyboard and still requires a lot of "typing ascii characters". ;-)
The "typing ascii characters" angle is a bit hyperbolic, I admit. But my intention is to get people to think about their software, not their personal experience of it.
BTW, there's nothing preventing you from using AI agents and staying in the flow state. If you want that, the universe is not stopping you. In fact, not dealing with the minutia of source code may well free us up and allow even greater flow experiences.
When I retire almost none of the software I produced during my career will still be in use, but I'll have memories of 40 years of work to live with until I pass away.
> In fact, not dealing with the minutia of source code may well free us up and allow even greater flow experiences.
You say minutia, but others say well organized notation and predictive systems. At least for me, writing code is as easy as writing English and with less effort.
> Do we really miss typing ascii characters into an editor?
Yes.
> Dare I say those stuck on nostalgia for pressing keys are demonstrating that they cared more about their own personal experience than about the outcome of their work
I care about the outcome, which is why I don't trust it to a fucking LLM
Fully agree. I never mentioned velocity or advocated for lower quality. In fact, this statement very well sums up my point: we should care about the thing we're producing, not our personal experience of coding it.
It may feel like you’re designing a bridge, but in reality you’re just asking a worker to do that too, and rubber stamping it when the drawings look reasonable.
One day you’ll be asked why it collapsed and realize you don’t actually understand it anymore.
Yeah honestly. I could easily architect the thing out but coding it is so boring. I love this new paradigm and sometimes the LLM even suggests a better architecture!
I would be perfectly fine with this new branch of "software development" if you guys didn't coopt our name and call yourselves programmers and engineers.
I did programming for 7 years (before I moved to management for 10). I’m definitely an engineer and LLMs at this point are not good enough where a non-engineer could come up with something solid out of vibe coding. That might not be true in 6 months.
Choose when and where and how you apply it and the sadness goes away. There's zero rule that you have to use an LLM in your workflow; even if your peers insist it's "stupid not to."
When I started to get this feeling recently (the sadness around the flow state being knocked off-axis), I started asking myself "what are you rushing toward? Do you really need to be working like this? Is this truly rational or just socially congruent?" YMMV but may be helpful for some.
I'm not sure how I feel (or should feel) when I read posts list this.
Here I am still coding (mostly) by hand.
While I also sometimes do chat with qwen or use an agent to save some time writting tests or yaml, or "implementing" a draft version of a change, I can't really understand this "the job is changed".
Do some companies in some countries force you to use these agents? Are they going to fire you because Jack or Jill push changes two (or more) times faster than you?
When ChatGPT came out, I was beyond excited. I had a tool to generate config files, and bounce ideas off of, and help unblock me when I got stupid arcane logs I didn't understand. That didn't feel existentially depressing at all; it was just glorified Google in my mind.
When the vibe coding tools like Claude and Codex came along, I got into this kind of dread. I'm sort of required to use them for work (we are "AI-first"...), but even if I weren't the tools are useful enough to me that I kind of feel like I have to use them because if I don't I'll be left in the dust.
And now it kind of feels like a lot of my job has been converted into babysitting interns. I don't get to write a lot of code by hand anymore, because most of what I do ends up being yelling at Codex to automate most of what I used to do. It's not all bad; I never got any enjoyment out of the initial bullshit of getting the initial project and configurations set up or futzing with configuration files, but I did get joy out of writing the actual implementation of the code, and now I don't get to do that much anymore.
A silver lining though; I do get to think in higher levels now, which is kind of fun. A lot of what I get to do now is write stuff in TLA+ and/or Mermaid (depending on the complexity and how much fancy concurrency I want to do), feed that into Claude, and get it to implement that. That part is fun, but I fear that I'm an outlier and that kind of programming won't catch on because engineers love to take the fucking idiotic position that they "don't need to do math to do programming".
Sure; for reference most of what I end up doing is concurrent and distributed systems, so that's more or less what I focus on. I don't know if this stuff translates to other domains.
Basically, when I have an idea for a project, I usually start drawing a diagram for how I think things should work. I usually draw it with draw.io first because it's easy and quick. After that, I usually translate this to Mermaid, primarily just to have a second draft, but also because having a text-based system makes it easy to copy and paste around, and that kind of stuff tends to translates well to state-machine based stuff that most distributed systems tend to morph into.
If what I'm working on is relatively simple, I feed the mermaid code into Claude or Codex, give it some surrounding context text about what I want, and get it to hack away.
For more complicated stuff, especially if I want to be more clever with concurrency, I will take my Mermaid diagram and manually convert it to TLA+. If I get stuck with the TLA+ translation, I will sometimes ask Claude for a bit of help but even then I almost always write 90+% of the TLA+ spec by hand.
After that it's basically the same. I feed the TLA+ spec, and provide context to it as well (e.g. how to translate a manual TLA+ mutex or channel implementation to whatever the language has built in).
Since I focus very heavily on the implementation, Claude and Codex generally don't have a huge issue with doing a one-to-one implementation.
I have heard of companies that are counting how much you spend on AI in your performance review (higher = better). I'm not at one of these, but upper management is pushing it hard, and I'm lucky that my team has a technical manager who understands the limitations. So I'm not using it either.
I'm in the same boat, I code mostly by hand. I really enjoy it. The OP seems to have given it up without a fight, and that doesn't make sense to me. Especially because with 30 years experience I'm guessing they have FU money and are probably better than most people using AI anyways. Keep doing what you love and "surrender to the flow".
Yes, companies will hire employees that are significantly faster if they can do some for around the same cost.
And they can - the current crop of tools provide undeniable productivity uplifts to developers, even if they’re only using the cheaper open weight models.
I was like you, using them for one shot things until January of this year. My token usage has roughly 100x’d since then, once I saw the value in the agentic loop.
The grateful dead were definitely my introduction to flow state. It wasn't just programming, it was using a MUD in the late 1980s. In my case, the Dark Star from Live/Dead (a long, extended jam that explores the outer reaches), I like a subset of the dead and phish for flow state programming but like the author find that modern work requires too many distractions to enter and stay in the flow state for a long time. In my case, large portions of the dead and phish don't work for me at all- the worst is when Phish is doing a great job and then suddenly they sort of lose the groove, or when they get stuck in repetitive loops.
Latency matters if you want programmers to be able to work interactively. We multitask because LLMs are too slow.
That's happened before, with people submitting batch jobs in the mainframe era, people working on big projects that take an hour to compile, or people waiting for code reviews.
However, even if the LLMs become fast, the coding agent will likely bottleneck on running tools. You will need to keep your tests fast, too.
Not Phish, but I have had music on in the background when coding for three decades at least now.
And not even necessarily when coding… I find that if there is no music playing, I will start a song in my head. That can get frustrating—the song loops, perhaps in fact just the chorus. So I turn on the external music and it seems to allow my brain to do other things.
The staccato framing is right. It's also describing the default posture, not the settled one.
That loop (open a session, ask a question, redirect, switch) happens before the spec discipline is tight enough to trust the result without watching. When you've actually delegated well, you're not context-switching constantly between sessions. You hand over a detailed spec, wait for the agent to finish, and evaluate. The watching gets compressed to the handoff and the return, not the whole run.
Most engineers haven't gotten there, because getting there is hard. The spec has to be good enough that the agent can hit an ambiguity without needing to ask, fill a gap without needing your correction, and still produce something worth using. That's a different kind of writing than what most engineers were doing before. It takes the kind of attention that's hard to find when you're also redirecting three other sessions.
The flow state question is the more interesting one. It doesn't disappear. It moves. The continuous-arc attention used to live at the implementation layer. Now it's available at the spec-writing layer. Most people haven't redesigned their work to find it there. The new deep-work mode runs from a high-level goal through a narrow, implementation-ready spec through a validation pass. That can be a continuous arc if you've structured the work to allow it. It just doesn't look like the old one, and the muscle memory for it doesn't exist yet.
Whether that's actually flow or just a different kind of concentration is a real question. The engineers that I know (myself included), who've found something like the old rhythm have moved most of their time/effort upstream in the process. The ones still in the redirect loop are usually starting with stories with the same information they always had in them, ran through an LLM to make them appear more tightly defined.
For me it’s the dead. Most days I’d listen to one or two shows from today in history on relisten.net/today.
I’m sysadmin not a dev but I also feel that managing an agent is a fundamentally different feeling than performing admin work like I used to. I used to build bridges out of software. I would learn how they work down to the nuts and bolts, and use them to make robust and occasionally clever solutions for our needs. Over time i was getting better at bash and powershell and regex and automating little tasks.
Gaining knowledge about kubernetes and building images and helm charts was some of the most fun I’ve had professionally. I found enjoyment and value in learning those things and enjoy knowing them for their own sake, much in the same way I enjoy being able to recite mechanics and knowledge about vanilla wow from memory. The knowledge is its own pursuit and obtaining it was fun and fulfilling.
Using AI is nothing like that. It’s not “fun” to me in any way. I don’t learn bash and powershell and templating. I don’t get to enjoy simple wins. AI does those all those for me.
I’m thinking of becoming an electrician. I can’t imagine babysitting an ai for another 30 years.
I completely understand the active/passive difference but I have a question for you that I think especially applies to sysadmin: isn't the knowledge you have and your continual learning just as important if what you're essentially doing now is using an LLM to help you find problems, propose solutions, and test the solutions?
Maybe it's even more important since you've gone from crafting your own solutions to reviewing the viablity of solutions from the LLM which occasionally are going to introduce new ideas, do things in a similar but slightly different way to how you would, or even be subtly wrong in ways that also exercise your ability to adapt and learn and nudge it in the right direction instead.
I don't really understand how the alternative could be true, in the sense of your knowledge not being relevant, as that would imply a couple of things that don't seem true as of now: that LLMs always produce optimal solutions in any given scenario, and (as a corollary) that sysadmin is entirely automated apart from some sort of thin beurocratic human approval/deployment layer.
In my experience so far I've found sysadmin type stuff to actually be what LLMs help me the least with, excluding text-in-text-out scripts, precisely because the setup often has to be so custom to the system, there's loads of side effects that are hard to communicate to the LLM so it oversimplifies and gets things wrong often.
So hopefully you can understand the implied challenge behind the question. I'm not saying your feeling of it being less fun is invalid. But I'm really curious about what you said specifically about the fun of applying knowledge and learning having been diminished. What do I have wrong?
So the biggest change is the day to day workflow change of just pumping out little powershell scripts for whatever I need and not building those scripting skills or performing the manual rote admin work. Rather than manually expand a disk in the hypervisor and then sshing in and expanding the partition and filesystem Claude spits out a one liner and I run it. While not challenging or particularly rewarding those trivial tasks are a soothing part of the constant flow of work that i didn’t dislike. Solid things in which i understood every aspect becomes a black box that does it for me and i have to reinterrogate or accept it, and in which it’ll never be worth the time to learn myself.
As for the deeper work, the most recent example I have is deploying an observability stack for our lab cluster. In the past i would have done way more upfront work on understanding the alternatives and the deployment. I would have known every line of config before pressing deploy including the ones I chose not to set. The why and how of everything. Now the most efficient way is just to tell Claude to stand up a monitoring stack in the manner that apps are already deployed on the cluster and iterate with Claude once it’s up and running. Why bother diving deep when you can be running a proof of concept in five minutes and then just see if it falls over.
Another part of it is that i really dislike learning through a prompt. I don’t like summaries, i like man pages. I don’t want to interrogate a result i want to provide a complete solution.
Part of it too is I’m fairly new to all this professionally. I’ve only been at it 5 years, though hobby much longer. Basically i can feel myself growing and getting better and now suddenly there’s this genius moron in my pocket that so much better than I ever dreamed I could be while simultaneously having huge shortfalls. It’s a just an instant paradigm shift and after the first week of glee it’s not been a positive feeling.
It’s not that my existing knowledge isn’t relevant at all, there’s still a huge base of networking and systems knowledge that’s necessary. It’s routinely surprised me in my career when i speak with a greybeard developer who is a an absolute wizard to me but dns/tcp etc are just a black box to them. Or when I talk to a friend new into the selfhosting hobby and i try to explain something to them that think is simple and i realize there’s 10 years of accumulated knowledge underneath that simple concept in order to actually grok it.
I don’t know if that answered your question, just an unstructured ramble. I think the heart of it is that googling and research and things being concrete and complete understandings of workings and interactions and variables and secrets and configs etc are giving way to a black box pumping out systems and it mostly working out of the box and often that is simply good enough for now and it’s on to the next thing.
appreciate the response. I understand better now, and agree with a lot of what you've said about kinda the mechanics of scripting stuff 'by hand' being just quite fun and satisfying but maybe no longer the expected way to do things, and so in some sense the sweet spot of that being something you can be paid to do is largely over. Interesting times for our profession.
It’s more than just syntax for me though, it was learning how to code in general, through bash. I was slowly building those skills and starting to apply those lessons when looking at product code. I could just devote that time to study instead but I’ve always learned by doing in my career. Now LLMSs do the doing.
Which parts are the grunt work depends on the individual. Personally, writing components was the most challenging, least grunt work job there was. Requirements and design was the next most fun, but not really very challenging. Writing glue and testing was grunt work--had to get done, but was mind-numbing.
As one of my friends put it, "Beej, writing software was never the goal."
I might counter that managing LLMs was never the goal, either, so we'll get to see where that leads.
I’ve been finding it difficult to hand things over to an LLM completely, and I wonder if this is part of it. I’ll let it help me organize at the beginning, and then I’ll have it come through and refactor or review, but the crafting part is where I want to spend my time. I love hitting tab but it seems like every time I do, I get this sensation like I’ve sort of time-warped into the future by a few seconds, and I wonder what I’d have written if the LLM hadn’t done it for me. At which point I’ll never know.
Great for generating volumes of output. Less great for going into a flow state and coming out with something that looks like I made it, something that I see my hand in
And yeah, maybe it’s because I never quite get into a flow state when I do it this way. Hmm
If the LLM behaved more like a programmer-in-flow-state and we could "hitch a ride" or "ride the wave" along, maybe we can recover some of this unbrokenness of attentivity.
We’re a long way from figuring out how to get flow state with agents. Maybe some sort of less stonerish upper-based master of the universe overseeing 100s of agents manic state.
I think there’d be a lot of demand from long time engineers that loved working in flow state to build tooling that encouraged flow. I think tokens/s needs to get like 10x faster first, because you’re going to be heading into a world where you are receiving very soft and non-distracting suggestions, probably at the periphery of your consciousness. Most will be thrown away.
I can kind of imagine a UI for this. I might experiment a little building something, but it will be by telling some agents to build it.. :)
The thing about flow state is that it is about being deeply focused on one thing. I’m not sure that orchestrating hundreds of agents will help—it’s not really about “CPU load”, at least, not for me.
I HAVE found another way to achieve flow state. Hand tool woodworking. Especially planing. There’s something about the rhythm of the tool, combined with the awareness of what the tool is doing, that makes the rest of the world evaporate, just like when coding. I could see myself doing that instead. Sadly there is even less call for slow woodworking than there is for slow coding.
> We’re a long way from figuring out how to get flow state with agents
its been the opposite for me. being able to create things with almost zero friction has lead me to have to slow down for my own sanity at times. I've had to setup "handcoding time" as a way to keep me touching virtual grass lest I completely lose myself building stuff at all hours.
literally thinking of taking up Haskell just to have a language that forces me to slow down and think in my off hours.
Yeah I agree — I don’t think I said what I’ve experienced very well — there is a different sort of state you can get in during a large multiagent session. But to me it’s more manic, not ‘smooth’ or peaceful. On a good day hand coding I will often feel groggy coming out of it, possibly subverbal, like I’ve recruited most of my brain to doing one thing for some time. I’ve never had a job that I could do that in, but I imagine the subset of engineers that do are pretty happy.
Tokens don’t matter if you have any slow deterministic processes. The bottleneck is already on testing infrastructure in the work I do. Screenshot tests are slow. Enormous codebases are slow. We have whole teams devoted to speeding these things up, so all I can do is get the agent to call the minimum set of commands necessary to check their latest iteration. This of course flies right in the face of the mountains of slow commands the harness tells it to use, resulting in chaos.
As a phish fan, this is an especially sad read. I wonder what new developers might feel, if they have the same sense of wonder and joy, listening to a brand new jam, working with an agent to create something, learning a thing or two about development, and being proud. I hope so.
This is a brilliant, obscure, high-fidelity indicator that speaks volumes (pun? pun intended) about the change and—critically—casts the change in unambiguous relief.
Programming does not exist or, rather, programming doesn't pay. Whatever this is—vibe coding, agentic software development—it's a new and different discipline, and it may be the only game in town [citation needed].
It's not even been a particularly gradual change. It's been a stark, totalizing turnover in the last 18 months. I don't know how long this era will last (maybe we'll discover a new sort of operational scurvy, and this movement will be mocked and scorned as a ludicrously anemic fad) but it'll leave a distinct layer of discoloration in the geological record.
I've never really been into Phish. Lately, I've been vibing out to the hyperactive chiptune groups Anamanaguchi and Toby Fox. Justice also makes my playlist, alongside more pathos-laden groups like The Glitch Mob and Moderat.
Hell, once I get this teams-of-teams jj-and-weave harness firmly in hand, I can pop into Agent-of-Empires and drop the needle on some Al Hirt—Music to Watch [Pulls] By.
FWIW my workflow is hours of intense research and design being constantly interrupted by agents finishing/choking on the last plan my previous research produced.
I think this fits well with Phish's isolated monoculture. I also started listening to Phish in the early 90s and have only seen 30 shows or so. Every time I go to a run it is very comfortable. There have been times I havn't seen them for a a good chunk of a decade and the shows feel the same. Eventhough jams have been part of their much of their history variation in musical style hasn't been. That leads to homoginization that makes for great vibe music.
I was about to argue with "Old timey", but yeah, I guess any band that has existed for over 40 years is old timey. I've been a fan since '96. I guess we're all old now.
The late 90s was the first time I'd heard of them. I was a university student at the time, and one of my friends was a foreign student, from America, and a fan. He also liked this other band called The Grateful Dead. I'd never heard of them either. They sucked too.
(Though, that said: many years later - i.e., quite recently - I listened to a bunch of their stuff on YouTube, while thinking back to my student days. And, actually, some of it is pretty good. At least, it was when they had pigpen singing.)
It explicitly says "the band". All the context in the first paragraphs is about the music. It absolutely clarifies. The rest of the article talks about listening to the music, watching them (or wanting to watch them) in concert or in tour.
Read the article. Or just look at the top search result for Phish. Stop being lazy.
Please don’t use ad hominem on HN. I did read the article and it wasn’t clear at all, as you can see others also faced that situation. You don’t have to engage if you think we are lazy. The aggressive tone is completely unwarranted for such a casual discussion
The top search results for "Phish" and "Down with Disease" show Phish the band. There is no ad hominem - writing a comment asking someone else to answer that question for you without doing that basic step is lazy.
The article says "Phish is a band". Verbatim. It talks about Phish and listening to music in every paragraph.
If people read this article and didn't get that, perhaps I should stop dismissing the news articles talking about the state of education and attention spans as hyperbole.
> I’d need to ask the interviewer to put Phish on....I could not, with any reliability, get into the zone without the music.
> All I ever wanted to do was listen to Phish and program
> I heard Phish for the first time at thirteen.
> Phish is a band that rewards you for staying in one place for a long time. The jams are long. The compositions unfold.
> The band would play in Hampton or Alpine Valley or wherever
> I have listened to Phish every day since I was fifteen...I have listened to certain shows so many times that I can sing the solos back, note by note,
> I tried to keep the music on. I’m writing this in the days after nine nights of Phish at the Sphere. Since I finished grad school and got a real job, I’ve gone to every show I could, every tour, every residency, making up for lost time. The music is more present in my life now than it has ever been. It isn’t what’s gone
I don't care if English is your first language - if anyone read the article they make the point that it is a band over and over and over. Explicitly.
I re read that paragraph a few times hoping that I missed something. Then gave up and went on to find the answer in comments. I thought the author was describing some feeling, or maybe some action. That "that to me" isn't clear with "that".
Long agent runs make such a difference. We focus a lot on new models and long context, but the bigger impact is in automatic verification.
I've been leaning in more on e2e test suites. They are slow, brittle and inefficient. But that's almost a feature. I can step away and come back an hour later, and use that time to think about bigger problems.
just make your claude code task completed notification sound play tweezer
feel this though. also how the era of debating like the best code styles and which new frontend framework is etc best that used to be fun to talk about feels like its coming to an end because no one really cares anymore as long as the bot can build the feature.
What is there to say except everyone is different and to each their own! I can't imagine wanting to listen to Phish for extended periods of time, muchless while working. But it's beautiful that someone else might. I am sorry for your loss original poster.
Our dormitory had two roommates, science majors, who always listened to rock music while studying. If their door was closed and you could hear music, they were lost in thought.
I never could study to music and much preferred studying in a quiet carrel in the university library.
Beautiful post. The job has totally changed in that last year and there are no guideposts. Maybe I should spend half of my work day in still zen meditation to maximise my ability to manage the agent swam for the rest of the day? Who even knows.
> The jams are built for one continuous arc of attention. The work is staccato.
Couldn't agree more. I'm personally okay with how engineering is changing. At the end of the day, the code is a means to an end for me. That said, the "queue" aspect of how software development is headed is so real. It's a different way of working, and I find the biggest challenge is staying engaged and tuned in while you might have agents take 30 seconds here, 2 minutes there, 5 minutes there, etc. It's easy to get distracted when waiting.
This sounds the closest to what I’d imagine Jordan was to winning. “All I ever wanted to do was listen to Phish and program”. As someone trying to find their why, this strikes home.
> OP, I sure it’s just a transition. AI will get faster, feedback loops shorter, and we will be back at a more traditional flow state.
You are right that it is a transitional state, but the end of the transition isn't that developers return to flow, it is that developers no longer have any reason to be employed.
The end state (assuming LLMs and other AI systems continue to improve at current rates or better) is one where everyone can produce any kind of digital content quickly with minimal toil.
Some people, who imagine themselves as modern Jobs-esque idea men of great taste like the idea of this but they aren't fully thinking about the consequences of easy content creation.
If you can get an LLM to produce a product of such clearly great design through your innate sense of taste, many someone elses can also point their agentic LLMs at your product, clone everything good about it very quickly and undersell you in a rapid race to zero value in a market that is a sea of competing noise.
As AI systems continue to improve at their core ability of mimicry there will be no moats in software anymore, for anyone.
I’m really disappointed that I wasn’t in the right headspace to name this post “the trick is to surrender to the flow”, the lyric from Phish’s The Lizards.
My brother still follows them across the country. Reading this and the comments gave me real insight into what longtime developers are feeling
I’m brand new to coding and have only done some AI-assisted work while learning. It blows my mind that people can do this at a high level without assistance - major respect
I think flow state can come back as the models get more sophisticated/mature and then more optimized for speed. It will not be the same type of work as before, but the flow of building and engineering at the speed of thought can happen again. You can see hints of this with the faster models today.
"This is engineering. I keep being told that" -- made me laugh out loud. We've gone from painting, to pointing to finished pieces in catalogues. Not to be utterly unhinged and tone-deaf but, there are other genres, such as jazz and screamo and noise that might suit the mood more aptly.
Am I the only one who doesn't get either side of this?
AI still sucks pretty bad at writing code. The only people I've ever known to need a "flow" state to write code are junior devs.
Everyone else is used to constant interruptions and has been through every layer of abstraction many many times. This is why those with experience find it so tempting to say this job can be automated away, but they forget how many gotchas there are, how they crop up, and how brittle all this crap always was. AI is actively making this problem worse.
Totally self inflicted btw. You can simply choose to not use agents in an autonomous way if you don't enjoy the workflow. Life is too short to do something you don't enjoy (employment aside ofc, but keep trying to change that).
For me it's Toto. Musically they are incredible; a supergroup of the best session musicians in the industry who worked on some of the most iconic albums of all time. But lyrically it's complete nonsense, so you can lose yourself in the music while still being focused on what you're doing, since the words are just basically gibberish.
I was around when cPanel and Wordpress broke through and saw professionals scoff at it and left it to others, who subsequently built the majority of web sites.
It's interesting to now see professionals making a similar move themselves when a new cPanel/Wordpress has arrived and wring their hands about how they 'get more done' but 'lost the craft'.
Inbetween I've come across quite a lot of scoffing about using XML to generate code in Java and C#, or just XML itself. Buying queries from a database corporation seems quite a bit weirder to me than designing a data schema, especially if both use cases are supposed to lead to some server boilerplate and an API on a database.
I feel his pain. I am more towards the opposite end of the spectrum [1]. I program to get things done. Usually, I don’t like programming. It’s too focused on one thing. Sometimes I like it though, precisely because it’s focused on one thing. But I love the things you can imagine and build with it. So for me, LLMs are amazing because I get to be an idea guy when I want but sometimes I can go deep, also when I want. I’ve kept up too, so I have the experience backing it up.
With that said, when I read this blog post. I feel the author’s pain. And it’s the first time that I emotionally get what the other side of the programming spectrum feels what it has lost. I feel sad about it. And because of it, I will also wonder about ways of bringing it back.
[1] not fully though. Something I can love coding by hand for months.
> So for me, LLMs are amazing because I get to be an idea guy
Enjoy this while it lasts, because very shortly LLMs are going to be fundamentally better at being the "idea guy" as well.
I had the same feeling reading this article today as I did reading the NYT article about the closing of Butter Ridge dairy farm. All that family wants to do is raise cows, name them, and milk them but it doesn't work as a business anymore because the factory farm machine took the viability away.
I feel the author deeply; this is coming for everyone across most of the economy.
> Enjoy this while it lasts, because very shortly LLMs are going to be fundamentally better at being the "idea guy" as well.
Maybe for the ideas that are so far from novel that there's a corpus of training data that could train an LLM to reproduce it. But for actually novel ideas, LLMs won't ever be much use. They can't interact with the world directly, they can only interact with text (or I guess bits).
And I guess that's not much of a distinction, as truly novel good ideas are very rare and you can go very far applying a good idea into a new domain. But at the edges where true novelty is required, LLMs will either hallucinate or guide you away from the edge.
No they won’t. And what you describe about the farm is just whitewashing marketing
So much comes down to timing - lucky author that got to experience it for 30 whole years whereas some of us had a scant few to enjoy building as it was, flow state, and some semblance of security. At least I made plans the last 5 years in anticipation of change, can't imagine the sheer dislocation of just entering the job market now.
Pros and cons. Some of the people who were lucky to enjoy those 30 years are also emotionally being hit the hardest right now, and if life threw a few curveballs at you along the way you don't necesarily have attained the sort of stability where you don't have to worry, either. Plus ageism can make it even harder to pivot.
I have programmer friends in their 40s to 60s who are seriously depressed currently (and 20 year olds worried for theirnl future perspectives, of course). Mental health is not just a young person's game.
I strangely feel quite lucky that I got more and more into electronics and hardware over time as I moved from web and desktop more and more into embedded/consumer electronics and companies who also employ mechanical and EE engineers. When I was younger I used to dumbly worry this meant giving something up (the purity of software approaches, etc.), but instead it made me consider myself an Engineer with a capital E and strive to learn the engineering method more generally, and learn so many other fields of the trade. It turns out this is a much more resilient identity than just Programmer and I recommend that approach.
It's funny that you mention EE. When I went back to school for CS in the early 2000s (I was a philosophy major... dumb), I was taking night classes with EEs who were fleeing to software. This was in the Boston area. A large contingent of my classmates worked at Avid. They started moving jobs out of the US and employees were told that if they switched to software they could remain employed. The mood was pretty grim, and these folks commented to me more than once that I was lucky that I loved writing software, because largely, they didn't.
Right, I also hear that sentiment pretty often. I don't regret having a software career and am very glad for my 25 years of C++ and many other things. I wouldn't want to be without that, and it probably did and does pay better.
However, it's pretty nice that these days I can also swing a semi-decent PCB, know my way around scopes and logic analyzers quite well, CAD something up for DFM in a number of processes from thermoplastics to machining, taught myself a fair bit of structural engineering, set up a FEM analysis correctly, etc. If only because it lets me bridge worlds and tie software and hardware together more effectively in the projects I'm in.
I cannot do any of these things as well as a seasoned veteran, but it has given me a broader appreciation of engineering overall and the commonalities between it all, to the point where I can also muster up leadership in engineering orgs more broadly and am not as hurt over the prospect that my pure programming skills might get devalued or diluted, or change.
For example, software engineers generally scoff at the perceived crustyness and lack of agility in classical mechanical engineering processes, but on the other hand mechanical engineering is far more experienced at defense-in-depth type approaches, dealing in components that have a failure rate to them and designing with error bars and safety factors, and I find some of that mindset has transferred quite naturally to engineering with our unreliable LLM friends at scale the past two years.
It takes a lot of the sting out if listening to Phish isn't your only move. Well, maybe not a lot, but at least it doesn't get so existential. Don't be a Programmer, be an Engineer. It's a lot easier to feel useful during a time of much doubt.
Everyone is talking about the agents part. I'm going to praise this post for describing it very clearly that some people, from a young age, don't need phases, growing up, trying things, figuring out, exploring the world, finding themselves.
Some people are just born something (engineers, in this case), and they're that something for life.
I always have a hard time explaining to "normal people" that such life is not boring at all, in fact, I can't remember a single time in my life where I was actually "bored".
The people who go through phases are also born to be something for life: adaptable, and used to change. Being good at doing exactly one thing your entire life a certain way has the potentially fatal flaw of having significant issue doing anything else.
I feel like I've lived 3-4 completely different lives so far, but the constant is the ability to adapt to the next one, and still find joy in it while you're there. "Survival is the ability to swim in strange water."
Personally, the AI tools have been transformative for work, but haven't affected how I work much. I have always coded as a team. I'd often do the largest and most complicated parts myself, but work (both at work work and my hobby work) has always been about passing things between colleagues based on what our strengths were and how much time we had.
The AI tools are another colleague. They work incredibly fast, and I do less coding myself now, but my goal was always to solve the problem, not the code itself. The AI tools do a great job most of the time, but they sometimes screw up and need more guidance or me to step in to fix the thing (usually a very small error compared to the whole). If they screw it all up, we might need to start from scratch, or I might need to just do it myself. But that's not most of the time. And then I figure out the thing missing in my process to move them in the right direction, and improve it.
I feel like any software developer used to collaborating with, training, and mentoring other developers knows exactly how to work with AI tools, and the only main difference is how much effort I put into being really nice about it the whole time.
One main thing has changed. Before I would handle the most complicated problems people were having trouble with myself. Now I allow AI to work on them, even if I know I'll have to fix it. The difference is that I care about the time, the strain, and the morale of my coworkers. AI is just getting paid for iterating tokens, so I don't have to feel bad about what I ask it to figure out for me.
> Some people are just born something (engineers, in this case), and they're that something for life.
Yes, and it's very not fun when your identity is being reshaped before your eyes in the matter of a couple years.
I wonder how many developers are going through real grief right now, while everybody else, lacking empathy, are just repeating "get a grip, it's just a tool" or "you better adapt or you're done".
Well, I know I have gone through these difficult emotions, and I choose now not to identify with my work, or at least my career any more. I certainly do not identify as what most people these days refer to as 'software engineer' any more.
This is a great reason not to identify too much with your work. I have enjoyed AI because it has reminded me that my real calling is art, and that I should be doing that at 8 pm, not coding
True, but if your true calling was coding, this change would be much harder to stomach
I don’t think anyone’s true calling is coding. That’s like saying you really like the act of writing, so much that you’d become a stenographer or a typist or something where you do zero higher level thinking and just absent mindedly press buttons.
Most people who are good at tech hate coding so much that they come up with elaborate abstractions so that they can avoid doing more of it.
I recommend you read the post because that's a really bad misunderstanding of the mindset, and like the comment at the top of this chain says, the post explains it well.
I read the post. I don’t agree with the “people are born to do one thing” mindset. There’s a lot of possibilities out there for everyone. I do identify with this OP fellow somewhat, except that I usually don’t code for fun on nights and weekends (also sunday code sesh can be fun)
Not sure if your 'enjoyed AI' is meant literally. I have escaped my existential crisis and found solace in art as well, simply because it's a very human method of self-expression and taking shortcuts to pain, effort and creativity by prompting LLMs is still frowned upon, at least by those that take art seriously.
The only way I can still enjoy programming now is if it's applied to artistic endeavours. I'm done with the soulless, cost-efficient software "engineering" (which by itself is a laughable proposition and a far cry from the high standards of other fields of engineering)
I do literally enjoy working with AI sometimes, other times it is hell. But sometimes coding by hand is hell, too, usually when working on other people’s extremely hacky and procedural code.
What I really enjoy is learning. I came back to computing after many years in another field, and was completely in love. Everything finally clicked, I'd spend hours reading everything I could, coding, trying things out, letting half-half-built projects pile up as I discovered new things.
AI has completely ruined this for me. Its boring having someone else do stuff for you. And worse, I feel I'm un-learning at a rapid rate.
The magic has gone, I'm not sure I want to be in this game in another five years.
Same for me
You don't know what your thing is until you try it.
Some people get lucky and the first thing they try turns out to be their thing.
Interesting that boring is the salient factor.
IMO assisted coding (auto complete style) has more flow state than the old days of getting stuck on obscure bugs (as satisfying as those were to crack).
Full agent coding however is the complete opposite, you’re in constant damage control of a junior who moves fast and breaks everything. They’ve got better but still do dumb mistakes.
lot of engineers are discovering firsthand what it’s like to manage a team of eager but useless employees. Not fun at all
The other thing is as far as I can tell, the power-tool style modes where you use AI to boost focus and do quick research is both much cheaper and, by the time you account for all the damage control/prompt tuning/other externalities, roughly as fast as full agentic.
I suspect with the prices going up, that realization is going to be pretty appealing.
Prices are not going up. DeepSeek V4 Pro is 5-10x cheaper than Claude 4.7.
As some of us have been predicting, model capability had already mostly plateaued, and the Chinese have and will continue to relentlessly push cost down. Chinese models will be used for 95% of things, with nation-native models for security/sovereignity-sensitive workloads. Eventually (5+ years from now), efficiency gains and hardware progress will make running local models the dominant way of doing things.
And yes, that puts the investors of Claude and OpenAI in quite a pickle.
I want the frontier on prem to be true but IMO not good enough yet unless async.
What started as all-you-can-eat $50 buffet has quietly become a $6k bill, frontier models that don’t ship your codebase to Beijing don’t come cheap anymore.
> I want the frontier on prem to be true
> that don’t ship your codebase to Beijing
DeepSeek is open weight, open source.
> DeepSeek is open weight, open source.
Sure, but I can't currently afford the hardware to run the frontier model.
Okay but that hardware doesn't have to be in Beijing, it can be in a Dutch data centre.
Yep. I’ve been using the Kagi Assistant with Kimi2 running out of providers in the US and Netherlands. The total monthly cost averages around $2.
It be great if more people drop good open models. But either way stuff like this is the future.
Yes, that’s basically what I’m talking about. Without subsidies, that’s a lot of incentive to use more efficient/open weight models, but also to use them in ways that are less compute intensive—fewer tokens, shorter reasoning chains, less nonstop, and generally tie up less hardware for a lot less time.
I agree, I also think there are long term costs to agentic coding that we won't see yet for a while, gradually and then suddenly.
It feels like nobody is even trying to iterate on the "power-tool style" usage of language models, everyone jumped straight to agents. It's not clear to me that removing a human from the loop is strictly more efficient though. Imagine an editor with an embedded language model (either running locally or using cheap cloud models) that is constantly churning, analysing code, reading debug logs, offering suggestions. Refactoring is not b̴e̴g̴g̴i̴n̴g̴ asking a chat interface to make changes, but structured refactors utilising the language models powered by the underlying ast representations to pull off much larger or involved refactors that surpass the abilities of current IDEs. Or doing codegen in your editor in a more structured (and thus repeatable) way compared to an agent spitting out code that you have to then review.
> IMO assisted coding (auto complete style) has more flow
I find this impossible to wrap my head around. Any time I autocomplete more than a single line, my flow is destroyed when I have to review what was just autocompleted. The only way I could remain in "flow" in this circumstance would be if I didn't review the code... I'm not ready to trust the LLMs that much yet
> IMO assisted coding (auto complete style) has more flow state than the old days of getting stuck on obscure bugs (as satisfying as those were to crack).
I strongly doubt that. Flow state happens when you can disassociate thinking and doing. Most of your brain is on the thinking while the doing lags behind. It’s like driving on a well known road. You don’t pay that much attention to the act of driving. Instead maybe you start planning what you want to do on your next stop
Using ai auto complete would be like your well known road, but turned into an obstacle course. You need your full focus on the road. Not that different from someone that is unfamiliar with the road, but not a flow state either. Full agentic coding is you being in the passenger seat with a driver known to be sometimes erratic. Less thinking but more worrying.
I feel like that even with regular autocomplete, I've often described it popping up as a mental speedbump.
I like completion, but not when it’s automatic. Usually the first thing I disable in an editor or IDE if I use it daily.
People are blown away when I tell them that, in the last 6 months, my job of coding has changed entirely, and that I now write very little code, but instead manage agents who write it. It is still engineering, and I still very much care what that code is, it's interfaces, how it interacts with the world, how it is tested, etc. etc., but it has taken me a while to get used to the idea of me not writing the code. I'm still not sure how I feel about it, although I am getting more done, and it has helped me keep better focus on "the big picture". That is tough to do when your day is spent in the weeds.
Really? When I've brought it up with anyone that doesn't code, I've found them to be totally disinterested in the topic.
It isn't about the code (they probably don't even know what that is), they focus on the fact that my daily job tasks have changed entirely (iow, I used to do 'something', 'what' doesn't matter, and now I dictate to an AI to do it). Most people can't fathom this unless they change to a completely different profession/job.
> I've found that nobody that doesn't code is at all interested in the topic.
This triple-negative is hard to parse.
I don't code (well, I write scripts for my own personal use and use Emacs), but I follow with interest these reports from software developers. Why? Because my profession is similar screen- and keyboard-based intellectual work, and what has come for devs will probably come for many other careers.
It's worth mentioning that one reason LLMs can clean up on coding tasks is because of the volumes of available data. Not only has the world produced copious volumes of code, they continue to produce copious volumes of code, and some code can even be generated synthetically (ie, not from an LLM, albeit at high cost).
Other domains are not like this. There will probably never be enough poetry out there to make an LLM do anything but be a poor imitation of a poet. This data is extremely hard to generate.
Coding with agents has also entered a positive feedback loop with million developers essentially paying to perform RL on frontier models.
Urm, isn't that a terrible example? LLMs are better at poetry and rhyming then most people trying their best...
Also you're thinking with an extremely short time horizon there.
All jobs which are centered around computers will be impacted the same way programming is today in the medium term. It's just a question of time until the data is gathered and tooling is adjusted. Because every company that's currently employing people to do something with software will start to use that data as training material, and a few years later they'll be swallowed along.
So that means eg bookkeeping is still safe... I sincerely doubt it'll be like that in 2035. It'll probably still be somewhat safe in 2030 though
I suspect most industries aren't quiet filled with people like the author though, most people treat their job as... Well, a job.
This is a wild take. Of all things, you think a large language model will be stumped by poetry?
I would wager everything I had that an AI could already win a blind poetry contest, especially if any effort was put into fine-tuning it. Unbelievable effort has been put into optimizing them for coding given the financials; next to none for poetry.
I disagree completely that what you describe is still “engineering.”
You are now a manager. In effect, you’ve consented to a career change.
I disagree. Even with agentic coding, you make engineering decisions.
Reading this, I felt a familiar kind of sadness. I have also felt some version of this recently: the sense of loss, and the question of whether I am still a “real programmer” if I am no longer writing code in the same way. There is a strange grief in letting go of a skill that once gave you pride.
But when I think about it from the author’s position, I may actually have been lucky. For this person, writing code may have been a way of life. In my case, I only started doing field work and using AI relatively recently, so I was able to adapt faster than I expected.
If your whole way of life changes, the shock must be much greater.
In contrast, I had no real status or social position to protect, so perhaps it was easier for me to let go. If I had tried to compete fairly and directly, I could not have beaten the experience and accumulated skill of veteran programmers.
Of course, my ability to write code was something I was somewhat proud of. Giving it up was painful, and it brought regret and a sense of inferiority. But at the same time, I also find myself thinking: “Was I really supposed to fight against veterans like this?”
Recently, this feels very similar to Durkheim’s concept of anomie. While reading this, I kept thinking about categories such as conformity, innovation, ritualism, retreatism, and rebellion. There are many points here that make me think.
If, in the future, coding changes again from today’s agent-based coding into some other form, what will happen to me then? By observing how senior programmers are reacting now, perhaps I can draw my own conclusions and prepare for that moment.
Right now, agent-based coding that depends on specific companies is dominant. But I think the current price of agentic coding is too cheap. At some point, when it becomes more expensive, local LLMs may become mainstream. If that happens, damaged or weakened code-writing ability may become necessary again.
So the question is: how should I prepare for that?
This was an interesting post.
I understand that commercial companies want to get as much value out of each developer as possible, I understand that managers want work to complete as quickly as possible. I can see why they are so excited about LLM tooling and the current increase in output.
This post makes a good point: managing LLM models isn't really the same thing as thinking hard about a particular problem, solving that problem and then concretizing it in code. If the work becomes managing models, I think we're going to see a pretty stark divergence between what people enjoy about developing software today and what the job is requiring. I'm not yet convinced LLM tooling will stick but, if it does, it makes me wonder what kind of person will be doing software development. Maybe some of same people and they find something else to enjoy about the job but I bet a lot of a different kind of person.
Personally, I am very uncomfortable with the idea that all software development might be mediated by LLM tooling and, as a consequence, require payment to a large corporate entity like Anthropic or Google. Hopefully some open source projects will remain open to accepting PRs from people, like the author of the OP, who enjoy working in that "flow" state. I enjoy writing software as a hobby (as well as at my job) but it looks like the hobby might become a larger source of personal fulfillment.
> I'm not yet convinced LLM tooling will stick but, if it does, it makes me wonder what kind of person will be doing software development. Maybe some of same people and they find something else to enjoy about the job but I bet a lot of a different kind of person.
TBH, I've often felt like a weirdo who enjoys "the wrong things" about software engineering.
Yep, I’m in this boat too.
I’ve given up trying to get that feeling back at work.
We were lucky for 20 years, now if we want to do it for craft it’s time to contribute to OpenBSD or something — with phish on, not for money.
Ditto. I programmed for 30 years with Underworld on. Listening to the beginning of Juanita/Kiteless instantly drops me into flow state. And it no longer fits if I am using an agent. I appear to be almost the same age as the writer of the post; just a little older. And I got that grocery store job.
So I am working on a personal OpenBSD-related project. I have no idea if it will ever see the light of day. It scratches an itch and I am having fun doing it, and I don’t use AI. It’s forcing me to read man pages carefully and OpenBSD source code (side note: the lack of comments in their code is almost a “fuck you”).
I wonder if I should try out some Phish.
To be fair, I often listen to Phish and Aphex Twin. Didn’t we all program to jungle in the 90s?
> Didn’t we all program to jungle in the 90s?
Still do!
Everything. Everything. Everything. :))
Not lager, lager, lager? ;)
> now if we want to do it for craft it’s time to contribute to OpenBSD or something
I think projects like these will end up being the highest-quality pieces of software out there.
If you found Phish twinned well with deep, focused work, you might enjoy grindcore for frequently interrupted work, since most of the songs are less than a minute long, like these: https://discordanceaxis.bandcamp.com/album/original-sound-ve...
I have a similar tendency. I doesn't really work for me to write code and listen to lyrics so EDM and grindcore, like carcass, works for me. Either there are few words or you can understand them.
Incredible comment, lmao
If you can't change the work, change the genre
Programming with Claude is still engineering. It is like designing a bridge, which remains engineering even when a worker pours the concrete instead of you.
In the past we were forced to pour the concrete ourselves. I understand how many of us enjoyed the sound and the smell of the concrete being poured. Myself, I’m happy to never get my hands dirty again, and focus on the actual engineering.
This is like the third bridge + concrete comparison I’ve seen in the past two days from a new account.
Did I miss something?
"justify your existence"
"You're absolutely correct. I don't just need to exist-I need to justify why. It's like builders with concrete..."
lmao, I think I read it here and then I proceeded to think I came up with it myself. I’m a fraud. (I put “concrete” into the search field and found it)
https://news.ycombinator.com/item?id=47996022
> the actual engineering
I was with you until the last three words. Craftsmanship in writing code is also “actual engineering”, it’s just not the engineering that people will pay a human to do now.
Chemical engineer here - I look at this similar to how a lot of professional engineers work, drafting gets passed off to eng techs, pfds get passed off to eng techs... The "actual engineering" is in the ideas, the applications, the reviews, and most importantly the final accountability.
Works great. No wonder China is eating our lunch.
Studied Electronic Engineering and have friends in Civil Engineering. By the time, you get to the actual implementation, the design has been tested so many times that errors are often material defects, not design defects. Engineering is about ironing out all the design defects and have all relevant unknowns be known.
Working with claude is closer to “grandma recipes” than any engineering or scientific practice.
Thanks for this perspective. Do we really miss typing ascii characters into an editor? That seems to me the least consequential and least interesting part of building software systems. Always has been.
Dare I say those stuck on nostalgia for pressing keys are demonstrating that they cared more about their own personal experience than about the outcome of their work? Now that coding is automated, we have to elevate our ambitions.
Ironically, Phish's music emerges from egoless expression (to paraphrase keyboardist Page McConnell). Giving up your own personal stake in the process is literally what brings something as beautiful as Phish's music into existence. We need to do the same with our software; give up the notion that "our" code is meaningful.
Dare I say those stuck on nostalgia for pressing keys are demonstrating that they cared more about their own personal experience than about the outcome of their work? Now that coding is automated, we have to elevate our ambitions.
YES. The beauty of programming is and always was that, first, you enjoyed it and, second, for some oddball reason you could actually get paid to do it. And one can't produce anything good unless you actually love working on it which means you want to put yourself working on it. The outcome might accidentally serve the one who pays for it but ultimately what did get the work finished was the sensation when you were reaching the point where you would finally tie things together and see everything you designed come to life and work together.
AI doesn't give you that personal involvement. We can do it but it's a different line of work and we care very little about what goes in and what comes out. We just do the grunt work of connecting the two ends. We're not for a fuck interested in elevating ambitions which is a word that relates to what is outside while all the good stuff comes from the inside.
I think concentrating on the physical act of typing on the keyboard is maybe taking it a little too far. The author of the OP talks more about holding a lot of the problem in their head and entering a "flow" state where they figure out a solution.
Most of my interaction with AI models and agents is still mediated by a keyboard and still requires a lot of "typing ascii characters". ;-)
The "typing ascii characters" angle is a bit hyperbolic, I admit. But my intention is to get people to think about their software, not their personal experience of it.
BTW, there's nothing preventing you from using AI agents and staying in the flow state. If you want that, the universe is not stopping you. In fact, not dealing with the minutia of source code may well free us up and allow even greater flow experiences.
When I retire almost none of the software I produced during my career will still be in use, but I'll have memories of 40 years of work to live with until I pass away.
> In fact, not dealing with the minutia of source code may well free us up and allow even greater flow experiences.
You say minutia, but others say well organized notation and predictive systems. At least for me, writing code is as easy as writing English and with less effort.
> Giving up your own personal stake in the process is literally what brings something as beautiful as Phish's music into existence
Unless Page McConnell is creating his music with AI these days, I don't think this point holds.
> Do we really miss typing ascii characters into an editor?
Yes.
> Dare I say those stuck on nostalgia for pressing keys are demonstrating that they cared more about their own personal experience than about the outcome of their work
I care about the outcome, which is why I don't trust it to a fucking LLM
Valuing quality over velocity is not selfish
> Valuing quality over velocity is not selfish
Fully agree. I never mentioned velocity or advocated for lower quality. In fact, this statement very well sums up my point: we should care about the thing we're producing, not our personal experience of coding it.
The product and the process are not orthogonal.
We should always care about our own subjectivity. If anything, subjectivity is too easy to discount in this day and age.
If you're arguing for using LLMs, you are effectively arguing in favor of sacrificing quality for speed as far as I'm concerned
In my experience anyone saying they're still producing high quality work while using LLMs is full of shit when the work is actually reviewed properly
It may feel like you’re designing a bridge, but in reality you’re just asking a worker to do that too, and rubber stamping it when the drawings look reasonable.
One day you’ll be asked why it collapsed and realize you don’t actually understand it anymore.
No, it’s management, not engineering.
It's like writing code with assistant Einstein who is 4 years old.
LLMs are the stereotypical 'idiot savant'
Yeah honestly. I could easily architect the thing out but coding it is so boring. I love this new paradigm and sometimes the LLM even suggests a better architecture!
I guess we are not the same.
I would be perfectly fine with this new branch of "software development" if you guys didn't coopt our name and call yourselves programmers and engineers.
I did programming for 7 years (before I moved to management for 10). I’m definitely an engineer and LLMs at this point are not good enough where a non-engineer could come up with something solid out of vibe coding. That might not be true in 6 months.
Choose when and where and how you apply it and the sadness goes away. There's zero rule that you have to use an LLM in your workflow; even if your peers insist it's "stupid not to."
When I started to get this feeling recently (the sadness around the flow state being knocked off-axis), I started asking myself "what are you rushing toward? Do you really need to be working like this? Is this truly rational or just socially congruent?" YMMV but may be helpful for some.
I'm not sure how I feel (or should feel) when I read posts list this.
Here I am still coding (mostly) by hand.
While I also sometimes do chat with qwen or use an agent to save some time writting tests or yaml, or "implementing" a draft version of a change, I can't really understand this "the job is changed".
Do some companies in some countries force you to use these agents? Are they going to fire you because Jack or Jill push changes two (or more) times faster than you?
When ChatGPT came out, I was beyond excited. I had a tool to generate config files, and bounce ideas off of, and help unblock me when I got stupid arcane logs I didn't understand. That didn't feel existentially depressing at all; it was just glorified Google in my mind.
When the vibe coding tools like Claude and Codex came along, I got into this kind of dread. I'm sort of required to use them for work (we are "AI-first"...), but even if I weren't the tools are useful enough to me that I kind of feel like I have to use them because if I don't I'll be left in the dust.
And now it kind of feels like a lot of my job has been converted into babysitting interns. I don't get to write a lot of code by hand anymore, because most of what I do ends up being yelling at Codex to automate most of what I used to do. It's not all bad; I never got any enjoyment out of the initial bullshit of getting the initial project and configurations set up or futzing with configuration files, but I did get joy out of writing the actual implementation of the code, and now I don't get to do that much anymore.
A silver lining though; I do get to think in higher levels now, which is kind of fun. A lot of what I get to do now is write stuff in TLA+ and/or Mermaid (depending on the complexity and how much fancy concurrency I want to do), feed that into Claude, and get it to implement that. That part is fun, but I fear that I'm an outlier and that kind of programming won't catch on because engineers love to take the fucking idiotic position that they "don't need to do math to do programming".
Can you describe how you work in this higher level? Sounds like scratches a similar itch that traditional programming offers.
Sure; for reference most of what I end up doing is concurrent and distributed systems, so that's more or less what I focus on. I don't know if this stuff translates to other domains.
Basically, when I have an idea for a project, I usually start drawing a diagram for how I think things should work. I usually draw it with draw.io first because it's easy and quick. After that, I usually translate this to Mermaid, primarily just to have a second draft, but also because having a text-based system makes it easy to copy and paste around, and that kind of stuff tends to translates well to state-machine based stuff that most distributed systems tend to morph into.
If what I'm working on is relatively simple, I feed the mermaid code into Claude or Codex, give it some surrounding context text about what I want, and get it to hack away.
For more complicated stuff, especially if I want to be more clever with concurrency, I will take my Mermaid diagram and manually convert it to TLA+. If I get stuck with the TLA+ translation, I will sometimes ask Claude for a bit of help but even then I almost always write 90+% of the TLA+ spec by hand.
After that it's basically the same. I feed the TLA+ spec, and provide context to it as well (e.g. how to translate a manual TLA+ mutex or channel implementation to whatever the language has built in).
Since I focus very heavily on the implementation, Claude and Codex generally don't have a huge issue with doing a one-to-one implementation.
I have heard of companies that are counting how much you spend on AI in your performance review (higher = better). I'm not at one of these, but upper management is pushing it hard, and I'm lucky that my team has a technical manager who understands the limitations. So I'm not using it either.
I'm in the same boat, I code mostly by hand. I really enjoy it. The OP seems to have given it up without a fight, and that doesn't make sense to me. Especially because with 30 years experience I'm guessing they have FU money and are probably better than most people using AI anyways. Keep doing what you love and "surrender to the flow".
Yes, companies will hire employees that are significantly faster if they can do some for around the same cost.
And they can - the current crop of tools provide undeniable productivity uplifts to developers, even if they’re only using the cheaper open weight models.
I was like you, using them for one shot things until January of this year. My token usage has roughly 100x’d since then, once I saw the value in the agentic loop.
>Yes, companies will hire employees that are significantly faster if they can do some for around the same cost.
And how do you test this when you hire? An interview where you have to implement something using LLM only?
Yes they will fire you. The sharks are circling those who just take tickets and implement what they say.
The grateful dead were definitely my introduction to flow state. It wasn't just programming, it was using a MUD in the late 1980s. In my case, the Dark Star from Live/Dead (a long, extended jam that explores the outer reaches), I like a subset of the dead and phish for flow state programming but like the author find that modern work requires too many distractions to enter and stay in the flow state for a long time. In my case, large portions of the dead and phish don't work for me at all- the worst is when Phish is doing a great job and then suddenly they sort of lose the groove, or when they get stuck in repetitive loops.
Latency matters if you want programmers to be able to work interactively. We multitask because LLMs are too slow.
That's happened before, with people submitting batch jobs in the mainframe era, people working on big projects that take an hour to compile, or people waiting for code reviews.
However, even if the LLMs become fast, the coding agent will likely bottleneck on running tools. You will need to keep your tests fast, too.
Not Phish, but I have had music on in the background when coding for three decades at least now.
And not even necessarily when coding… I find that if there is no music playing, I will start a song in my head. That can get frustrating—the song loops, perhaps in fact just the chorus. So I turn on the external music and it seems to allow my brain to do other things.
The staccato framing is right. It's also describing the default posture, not the settled one.
That loop (open a session, ask a question, redirect, switch) happens before the spec discipline is tight enough to trust the result without watching. When you've actually delegated well, you're not context-switching constantly between sessions. You hand over a detailed spec, wait for the agent to finish, and evaluate. The watching gets compressed to the handoff and the return, not the whole run.
Most engineers haven't gotten there, because getting there is hard. The spec has to be good enough that the agent can hit an ambiguity without needing to ask, fill a gap without needing your correction, and still produce something worth using. That's a different kind of writing than what most engineers were doing before. It takes the kind of attention that's hard to find when you're also redirecting three other sessions.
The flow state question is the more interesting one. It doesn't disappear. It moves. The continuous-arc attention used to live at the implementation layer. Now it's available at the spec-writing layer. Most people haven't redesigned their work to find it there. The new deep-work mode runs from a high-level goal through a narrow, implementation-ready spec through a validation pass. That can be a continuous arc if you've structured the work to allow it. It just doesn't look like the old one, and the muscle memory for it doesn't exist yet.
Whether that's actually flow or just a different kind of concentration is a real question. The engineers that I know (myself included), who've found something like the old rhythm have moved most of their time/effort upstream in the process. The ones still in the redirect loop are usually starting with stories with the same information they always had in them, ran through an LLM to make them appear more tightly defined.
For me it’s the dead. Most days I’d listen to one or two shows from today in history on relisten.net/today.
I’m sysadmin not a dev but I also feel that managing an agent is a fundamentally different feeling than performing admin work like I used to. I used to build bridges out of software. I would learn how they work down to the nuts and bolts, and use them to make robust and occasionally clever solutions for our needs. Over time i was getting better at bash and powershell and regex and automating little tasks.
Gaining knowledge about kubernetes and building images and helm charts was some of the most fun I’ve had professionally. I found enjoyment and value in learning those things and enjoy knowing them for their own sake, much in the same way I enjoy being able to recite mechanics and knowledge about vanilla wow from memory. The knowledge is its own pursuit and obtaining it was fun and fulfilling.
Using AI is nothing like that. It’s not “fun” to me in any way. I don’t learn bash and powershell and templating. I don’t get to enjoy simple wins. AI does those all those for me.
I’m thinking of becoming an electrician. I can’t imagine babysitting an ai for another 30 years.
I completely understand the active/passive difference but I have a question for you that I think especially applies to sysadmin: isn't the knowledge you have and your continual learning just as important if what you're essentially doing now is using an LLM to help you find problems, propose solutions, and test the solutions?
Maybe it's even more important since you've gone from crafting your own solutions to reviewing the viablity of solutions from the LLM which occasionally are going to introduce new ideas, do things in a similar but slightly different way to how you would, or even be subtly wrong in ways that also exercise your ability to adapt and learn and nudge it in the right direction instead.
I don't really understand how the alternative could be true, in the sense of your knowledge not being relevant, as that would imply a couple of things that don't seem true as of now: that LLMs always produce optimal solutions in any given scenario, and (as a corollary) that sysadmin is entirely automated apart from some sort of thin beurocratic human approval/deployment layer.
In my experience so far I've found sysadmin type stuff to actually be what LLMs help me the least with, excluding text-in-text-out scripts, precisely because the setup often has to be so custom to the system, there's loads of side effects that are hard to communicate to the LLM so it oversimplifies and gets things wrong often.
So hopefully you can understand the implied challenge behind the question. I'm not saying your feeling of it being less fun is invalid. But I'm really curious about what you said specifically about the fun of applying knowledge and learning having been diminished. What do I have wrong?
So the biggest change is the day to day workflow change of just pumping out little powershell scripts for whatever I need and not building those scripting skills or performing the manual rote admin work. Rather than manually expand a disk in the hypervisor and then sshing in and expanding the partition and filesystem Claude spits out a one liner and I run it. While not challenging or particularly rewarding those trivial tasks are a soothing part of the constant flow of work that i didn’t dislike. Solid things in which i understood every aspect becomes a black box that does it for me and i have to reinterrogate or accept it, and in which it’ll never be worth the time to learn myself.
As for the deeper work, the most recent example I have is deploying an observability stack for our lab cluster. In the past i would have done way more upfront work on understanding the alternatives and the deployment. I would have known every line of config before pressing deploy including the ones I chose not to set. The why and how of everything. Now the most efficient way is just to tell Claude to stand up a monitoring stack in the manner that apps are already deployed on the cluster and iterate with Claude once it’s up and running. Why bother diving deep when you can be running a proof of concept in five minutes and then just see if it falls over.
Another part of it is that i really dislike learning through a prompt. I don’t like summaries, i like man pages. I don’t want to interrogate a result i want to provide a complete solution.
Part of it too is I’m fairly new to all this professionally. I’ve only been at it 5 years, though hobby much longer. Basically i can feel myself growing and getting better and now suddenly there’s this genius moron in my pocket that so much better than I ever dreamed I could be while simultaneously having huge shortfalls. It’s a just an instant paradigm shift and after the first week of glee it’s not been a positive feeling.
It’s not that my existing knowledge isn’t relevant at all, there’s still a huge base of networking and systems knowledge that’s necessary. It’s routinely surprised me in my career when i speak with a greybeard developer who is a an absolute wizard to me but dns/tcp etc are just a black box to them. Or when I talk to a friend new into the selfhosting hobby and i try to explain something to them that think is simple and i realize there’s 10 years of accumulated knowledge underneath that simple concept in order to actually grok it.
I don’t know if that answered your question, just an unstructured ramble. I think the heart of it is that googling and research and things being concrete and complete understandings of workings and interactions and variables and secrets and configs etc are giving way to a black box pumping out systems and it mostly working out of the box and often that is simply good enough for now and it’s on to the next thing.
appreciate the response. I understand better now, and agree with a lot of what you've said about kinda the mechanics of scripting stuff 'by hand' being just quite fun and satisfying but maybe no longer the expected way to do things, and so in some sense the sweet spot of that being something you can be paid to do is largely over. Interesting times for our profession.
Yes certainly, I used to have some sense of the future now it’s hard to see past a year.
I've learnt more, and more quickly, with LLMs than before them.
Not syntax, no. Bash and Powershell are syntex. But tradeoffs, concepts, understanding new domains I didn't know about.
It’s more than just syntax for me though, it was learning how to code in general, through bash. I was slowly building those skills and starting to apply those lessons when looking at product code. I could just devote that time to study instead but I’ve always learned by doing in my career. Now LLMSs do the doing.
What is the "grunt work"? I think there's a difference of opinion with the commenters. :)
The work is, among other things:
* Writing components * Writing glue * Requirements gathering * Architectural design * Testing
Which parts are the grunt work depends on the individual. Personally, writing components was the most challenging, least grunt work job there was. Requirements and design was the next most fun, but not really very challenging. Writing glue and testing was grunt work--had to get done, but was mind-numbing.
As one of my friends put it, "Beej, writing software was never the goal."
I might counter that managing LLMs was never the goal, either, so we'll get to see where that leads.
I’ve been finding it difficult to hand things over to an LLM completely, and I wonder if this is part of it. I’ll let it help me organize at the beginning, and then I’ll have it come through and refactor or review, but the crafting part is where I want to spend my time. I love hitting tab but it seems like every time I do, I get this sensation like I’ve sort of time-warped into the future by a few seconds, and I wonder what I’d have written if the LLM hadn’t done it for me. At which point I’ll never know.
Great for generating volumes of output. Less great for going into a flow state and coming out with something that looks like I made it, something that I see my hand in
And yeah, maybe it’s because I never quite get into a flow state when I do it this way. Hmm
If the LLM behaved more like a programmer-in-flow-state and we could "hitch a ride" or "ride the wave" along, maybe we can recover some of this unbrokenness of attentivity.
you’re just complaining about unwittingly being promoted from an individual contributor to a manager.
"the higher up the corporate hierarchy, the less music"
We’re a long way from figuring out how to get flow state with agents. Maybe some sort of less stonerish upper-based master of the universe overseeing 100s of agents manic state.
I think there’d be a lot of demand from long time engineers that loved working in flow state to build tooling that encouraged flow. I think tokens/s needs to get like 10x faster first, because you’re going to be heading into a world where you are receiving very soft and non-distracting suggestions, probably at the periphery of your consciousness. Most will be thrown away.
I can kind of imagine a UI for this. I might experiment a little building something, but it will be by telling some agents to build it.. :)
The thing about flow state is that it is about being deeply focused on one thing. I’m not sure that orchestrating hundreds of agents will help—it’s not really about “CPU load”, at least, not for me.
I HAVE found another way to achieve flow state. Hand tool woodworking. Especially planing. There’s something about the rhythm of the tool, combined with the awareness of what the tool is doing, that makes the rest of the world evaporate, just like when coding. I could see myself doing that instead. Sadly there is even less call for slow woodworking than there is for slow coding.
> We’re a long way from figuring out how to get flow state with agents
its been the opposite for me. being able to create things with almost zero friction has lead me to have to slow down for my own sanity at times. I've had to setup "handcoding time" as a way to keep me touching virtual grass lest I completely lose myself building stuff at all hours.
literally thinking of taking up Haskell just to have a language that forces me to slow down and think in my off hours.
Yeah I agree — I don’t think I said what I’ve experienced very well — there is a different sort of state you can get in during a large multiagent session. But to me it’s more manic, not ‘smooth’ or peaceful. On a good day hand coding I will often feel groggy coming out of it, possibly subverbal, like I’ve recruited most of my brain to doing one thing for some time. I’ve never had a job that I could do that in, but I imagine the subset of engineers that do are pretty happy.
Tokens don’t matter if you have any slow deterministic processes. The bottleneck is already on testing infrastructure in the work I do. Screenshot tests are slow. Enormous codebases are slow. We have whole teams devoted to speeding these things up, so all I can do is get the agent to call the minimum set of commands necessary to check their latest iteration. This of course flies right in the face of the mountains of slow commands the harness tells it to use, resulting in chaos.
As a phish fan, this is an especially sad read. I wonder what new developers might feel, if they have the same sense of wonder and joy, listening to a brand new jam, working with an agent to create something, learning a thing or two about development, and being proud. I hope so.
This is a brilliant, obscure, high-fidelity indicator that speaks volumes (pun? pun intended) about the change and—critically—casts the change in unambiguous relief.
Programming does not exist or, rather, programming doesn't pay. Whatever this is—vibe coding, agentic software development—it's a new and different discipline, and it may be the only game in town [citation needed].
It's not even been a particularly gradual change. It's been a stark, totalizing turnover in the last 18 months. I don't know how long this era will last (maybe we'll discover a new sort of operational scurvy, and this movement will be mocked and scorned as a ludicrously anemic fad) but it'll leave a distinct layer of discoloration in the geological record.
I've never really been into Phish. Lately, I've been vibing out to the hyperactive chiptune groups Anamanaguchi and Toby Fox. Justice also makes my playlist, alongside more pathos-laden groups like The Glitch Mob and Moderat.
Hell, once I get this teams-of-teams jj-and-weave harness firmly in hand, I can pop into Agent-of-Empires and drop the needle on some Al Hirt—Music to Watch [Pulls] By.
FWIW my workflow is hours of intense research and design being constantly interrupted by agents finishing/choking on the last plan my previous research produced.
I think this fits well with Phish's isolated monoculture. I also started listening to Phish in the early 90s and have only seen 30 shows or so. Every time I go to a run it is very comfortable. There have been times I havn't seen them for a a good chunk of a decade and the shows feel the same. Eventhough jams have been part of their much of their history variation in musical style hasn't been. That leads to homoginization that makes for great vibe music.
Sad and lovely and speaks to how I’ve been feeling as well. I’ve been writing code since I was a boy and now I’m nearing retirement. I also miss flow.
What is Phish?
Phish is an American Rock Band. https://en.wikipedia.org/wiki/Phish
Old timey American guitar noodler band.
I was about to argue with "Old timey", but yeah, I guess any band that has existed for over 40 years is old timey. I've been a fan since '96. I guess we're all old now.
The late 90s was the first time I'd heard of them. I was a university student at the time, and one of my friends was a foreign student, from America, and a fan. He also liked this other band called The Grateful Dead. I'd never heard of them either. They sucked too.
(Though, that said: many years later - i.e., quite recently - I listened to a bunch of their stuff on YouTube, while thinking back to my student days. And, actually, some of it is pretty good. At least, it was when they had pigpen singing.)
(Also: if you are a connoisseur of almostbutnotquiteinsanepersonposting, allow me to highly recommend The Grateful Dead's sound engineer's series of posts about his diet: https://justmeat.co/archives/active-low-carber-forum-posts/)
Anyway, if it's from the last century, it's old timey to me.
now - 29 years = 1996: Phish - Billy Breathes
1996 - 29 years = 1967: The Doors - Light My Fire
It’s a full band improvisation when at its best not just guitar noodling thank you very much
Sentences 5 and 6 of the article, top of the second paragraph:
> That move is Phish fans in miniature. Someone cared enough about the song and the bit that they rebuilt a piece of pop culture around the band.
That doesn’t really clarify
It explicitly says "the band". All the context in the first paragraphs is about the music. It absolutely clarifies. The rest of the article talks about listening to the music, watching them (or wanting to watch them) in concert or in tour.
Read the article. Or just look at the top search result for Phish. Stop being lazy.
I literally read the entire article without understanding that Phish is a band.
Call me lazy if you will. You could also have chosen to just say they're a band and recommend me a record to listen to.
Something something 1 in 10'000.
Asking others to explain things without doing a single search of your own is lazy, as is not reading the article:
> Phish is a band that rewards you for staying in one place for a long time. The jams are long. The compositions unfold.
"Phish is a band". Literally. Verbatim.
Coworkers who do that get told not to, and if they do it a second time their boss gets told to talk to them about it.
Please don’t use ad hominem on HN. I did read the article and it wasn’t clear at all, as you can see others also faced that situation. You don’t have to engage if you think we are lazy. The aggressive tone is completely unwarranted for such a casual discussion
The top search results for "Phish" and "Down with Disease" show Phish the band. There is no ad hominem - writing a comment asking someone else to answer that question for you without doing that basic step is lazy.
The article says "Phish is a band". Verbatim. It talks about Phish and listening to music in every paragraph.
If people read this article and didn't get that, perhaps I should stop dismissing the news articles talking about the state of education and attention spans as hyperbole.
> I’d need to ask the interviewer to put Phish on....I could not, with any reliability, get into the zone without the music.
> All I ever wanted to do was listen to Phish and program
> I heard Phish for the first time at thirteen.
> Phish is a band that rewards you for staying in one place for a long time. The jams are long. The compositions unfold.
> The band would play in Hampton or Alpine Valley or wherever
> I have listened to Phish every day since I was fifteen...I have listened to certain shows so many times that I can sing the solos back, note by note,
> I tried to keep the music on. I’m writing this in the days after nine nights of Phish at the Sphere. Since I finished grad school and got a real job, I’ve gone to every show I could, every tour, every residency, making up for lost time. The music is more present in my life now than it has ever been. It isn’t what’s gone
I don't care if English is your first language - if anyone read the article they make the point that it is a band over and over and over. Explicitly.
Stop being lazy in this context is by definition ad hominem. I won’t engage with the rest though.
The whole rest of the sentence implies they're talking about something else based on the band, not the band itself.
I re read that paragraph a few times hoping that I missed something. Then gave up and went on to find the answer in comments. I thought the author was describing some feeling, or maybe some action. That "that to me" isn't clear with "that".
That was merely the first occurrence. Elsewhere the article says verbatim "Phish is a band".
It's who the dead heads follow around now.
As well as Billy Strings, Goose, King Gizzard & the Lizard Wizard, Sturgill Simpson, John Mayer…big scene these days
Long agent runs make such a difference. We focus a lot on new models and long context, but the bigger impact is in automatic verification.
I've been leaning in more on e2e test suites. They are slow, brittle and inefficient. But that's almost a feature. I can step away and come back an hour later, and use that time to think about bigger problems.
just make your claude code task completed notification sound play tweezer
feel this though. also how the era of debating like the best code styles and which new frontend framework is etc best that used to be fun to talk about feels like its coming to an end because no one really cares anymore as long as the bot can build the feature.
What is there to say except everyone is different and to each their own! I can't imagine wanting to listen to Phish for extended periods of time, muchless while working. But it's beautiful that someone else might. I am sorry for your loss original poster.
Same here. I used to listen to techno and could just disappear into the work. Those days are gone now.
Try this https://soundcloud.com/jochempaap/07-01-2010-speedy-j-bunker...
This was a colleague favourite for flow state time
Our dormitory had two roommates, science majors, who always listened to rock music while studying. If their door was closed and you could hear music, they were lost in thought.
I never could study to music and much preferred studying in a quiet carrel in the university library.
Beautiful post. The job has totally changed in that last year and there are no guideposts. Maybe I should spend half of my work day in still zen meditation to maximise my ability to manage the agent swam for the rest of the day? Who even knows.
> I’m sad.
Simple solution: keep going what you've been doing. Open up https://relisten.net/ and keep jamming, you'll probably be fine.
> The jams are built for one continuous arc of attention. The work is staccato.
Couldn't agree more. I'm personally okay with how engineering is changing. At the end of the day, the code is a means to an end for me. That said, the "queue" aspect of how software development is headed is so real. It's a different way of working, and I find the biggest challenge is staying engaged and tuned in while you might have agents take 30 seconds here, 2 minutes there, 5 minutes there, etc. It's easy to get distracted when waiting.
This sounds the closest to what I’d imagine Jordan was to winning. “All I ever wanted to do was listen to Phish and program”. As someone trying to find their why, this strikes home.
> Phish
Thought it was an early AI or something.
OP, I sure it’s just a transition. AI will get faster, feedback loops shorter, and we will be back at a more traditional flow state.
> OP, I sure it’s just a transition. AI will get faster, feedback loops shorter, and we will be back at a more traditional flow state.
You are right that it is a transitional state, but the end of the transition isn't that developers return to flow, it is that developers no longer have any reason to be employed.
The end state (assuming LLMs and other AI systems continue to improve at current rates or better) is one where everyone can produce any kind of digital content quickly with minimal toil.
Some people, who imagine themselves as modern Jobs-esque idea men of great taste like the idea of this but they aren't fully thinking about the consequences of easy content creation.
If you can get an LLM to produce a product of such clearly great design through your innate sense of taste, many someone elses can also point their agentic LLMs at your product, clone everything good about it very quickly and undersell you in a rapid race to zero value in a market that is a sea of competing noise.
As AI systems continue to improve at their core ability of mimicry there will be no moats in software anymore, for anyone.
Beautifully written and while I'm not a Phish fan per se, this otherwise summarized exactly how I've been feeling this year.
Phish? It turns out to be a band: https://phish.com/
Same. Used to listen 100k+ minutes/year in Spotify, most of it being Racionais MC’s while working and it’s completely gone.
This is me. Different music, same deal. Finding you’re not as alone as you thought you were is what the internet is for.
Oddly enough i have a similar bit about electronica, but its a fine line between productive programming and distraction
Hi, author here!
I’m really disappointed that I wasn’t in the right headspace to name this post “the trick is to surrender to the flow”, the lyric from Phish’s The Lizards.
But, you know, hindsight.
That would’ve been a great title!
My brother still follows them across the country. Reading this and the comments gave me real insight into what longtime developers are feeling
I’m brand new to coding and have only done some AI-assisted work while learning. It blows my mind that people can do this at a high level without assistance - major respect
I hope you find that flow again
I think flow state can come back as the models get more sophisticated/mature and then more optimized for speed. It will not be the same type of work as before, but the flow of building and engineering at the speed of thought can happen again. You can see hints of this with the faster models today.
This captures a lot of how I've been feeling lately. Thanks for sharing
I wonder if he was listening to Phish while playing Doom.
Been programming since I was 14 - now 44.
The flow state is gone. Sadly
"This is engineering. I keep being told that" -- made me laugh out loud. We've gone from painting, to pointing to finished pieces in catalogues. Not to be utterly unhinged and tone-deaf but, there are other genres, such as jazz and screamo and noise that might suit the mood more aptly.
Am I the only one who doesn't get either side of this?
AI still sucks pretty bad at writing code. The only people I've ever known to need a "flow" state to write code are junior devs.
Everyone else is used to constant interruptions and has been through every layer of abstraction many many times. This is why those with experience find it so tempting to say this job can be automated away, but they forget how many gotchas there are, how they crop up, and how brittle all this crap always was. AI is actively making this problem worse.
Totally self inflicted btw. You can simply choose to not use agents in an autonomous way if you don't enjoy the workflow. Life is too short to do something you don't enjoy (employment aside ofc, but keep trying to change that).
I have never understood how people can program while listening to music.
Where you get your satisfaction matters. Is it in the end product? Or the journey producing it?
I've been writing code for 30 years. I welcome AI. It brings different challenges and can reduce the time needed for the end product.
absolutely the journey.
Can we get the link to that TikTok? lol
Also what do you think about goose
Now the echo of my system prompts are bouncing round the room.
edit: Bouncing around the room is one of their hit songs. Give it a listen.
For me it's Toto. Musically they are incredible; a supergroup of the best session musicians in the industry who worked on some of the most iconic albums of all time. But lyrically it's complete nonsense, so you can lose yourself in the music while still being focused on what you're doing, since the words are just basically gibberish.
I was around when cPanel and Wordpress broke through and saw professionals scoff at it and left it to others, who subsequently built the majority of web sites.
It's interesting to now see professionals making a similar move themselves when a new cPanel/Wordpress has arrived and wring their hands about how they 'get more done' but 'lost the craft'.
Inbetween I've come across quite a lot of scoffing about using XML to generate code in Java and C#, or just XML itself. Buying queries from a database corporation seems quite a bit weirder to me than designing a data schema, especially if both use cases are supposed to lead to some server boilerplate and an API on a database.