Was very moved by this personal essay about generative AI degrading flow state. It's relatable for me even though I'm not a professional programmer. It's not necessarily AI specifically (although...
Was very moved by this personal essay about generative AI degrading flow state.
It's relatable for me even though I'm not a professional programmer. It's not necessarily AI specifically (although that's definitely a significant factor for me), but also things like rapid context-switching and algorithm-controlled feeds (which I both consume content from and produce content for).
I've been a professional visual artist most of my adult life, but I've been struggling with getting into a flow state with that recently. I'm currently learning game development and I really enjoy coding (even if I'm not good at it and make a lot of spaghetti!), so it's been easier for me to get into a flow state with that at the moment.
Excerpts below:
I used to make a joke that if I ever had to interview for a new job, I’d need to ask the interviewer to put Phish on so I could actually program for them. I’d say it as a joke, because saying it straight would have made it sound deranged. But it wasn’t a joke. After three decades, the cue and the state had fused. I could not, with any reliability, get into the zone without the music. The conditioning was complete and I knew it.
...
Since January, the work has changed.
I don’t really write code anymore. The main thing now is managing agents. I open a session, ask a question, redirect, switch to a different one, check on a merge, review what came back, send it back for changes, switch again. The day is a queue. Things finish at different times and require different responses, and the responses are short and the contexts are constantly different.
This is engineering. I keep being told that. It is engineering and it is the future and it is more leveraged than what I used to do. All of that is probably true. But it is not the work I have been doing for thirty years. The shape of it is different. The rhythm is different. The way it sits in the day is different.
...
The music is more present in my life now than it has ever been. It isn’t what’s gone.
But the music is out of phase with the work. The jams are built for one continuous arc of attention. The work is staccato. I’ll be three minutes into a song and I will have already context-switched four times. The song is happening and the work is happening and they’re no longer happening together. They’re parallel, but they no longer touch.
I’m sad. I don’t get into that state anymore. I don’t know how to be honest about this without sounding like I am complaining about progress, but I can’t pretend that something hasn’t been taken. The flow state I had for thirty years is not part of my workday now. The creativity that lived inside it is not there either. I do useful things. I do not feel what I used to feel while doing them.
The discussion about this on Hacker News may also be of interest, since there are others sharing differing perspectives. For some, coding with AI has improved flow state since they prefer working rapidly with ideas / design / architecture.
This reads like working on an assembly line or in a warehouse, but instead it's on a pc. My worst job was order picking in my teens. Every menial task is in a queue waiting to be picked up. No...
The main thing now is managing agents. I open a session, ask a question, redirect, switch to a different one, check on a merge, review what came back, send it back for changes, switch again. The day is a queue.
This reads like working on an assembly line or in a warehouse, but instead it's on a pc.
My worst job was order picking in my teens. Every menial task is in a queue waiting to be picked up. No meaningful variation. You work through it until you clock out. Next day however, another queue piled up.
There's little more soul crushing than monotony like that.
I don't think it's much like working in a warehouse. It might be like a support job (responding to trouble tickets) except that you never get abuse from a coding agent, you can rely on some...
I don't think it's much like working in a warehouse. It might be like a support job (responding to trouble tickets) except that you never get abuse from a coding agent, you can rely on some baseline intelligence even if it does screw up sometimes, and you're the one giving orders. (Or suggestions, which is how I usually phrase it.)
The multitasking is because LLM's are often too slow to wait around for; if they were faster then there wouldn't be much point switching tasks.
Since I'm just coding for fun, I multitask with something else (like checking Tildes or doing something else entirely) rather than trying to multitask programming.
Oh no I'm aware right now it's nothing like working in a warehouse, just that description is eerily similar. Nevertheless, a future where all a programmer does is verify code written by AI isn't...
Oh no I'm aware right now it's nothing like working in a warehouse, just that description is eerily similar. Nevertheless, a future where all a programmer does is verify code written by AI isn't too far off menial work.
I believe you! Your comment reminded me of this personal essay by Aella where she talks about how she started on this job feeling so excited and eager to learn and improve, but it just turned out...
I hated the work. It was a horrible no-good kind of work. You couldn’t zone out and just do autopilot, because you had to follow a set of instructions with slight variations, that change a little each time. You had to pay attention. But the attention you paid was not interesting; it was ongoing, it was relentless, it was stupid. My mind wasn’t mine, it was just barely pulled over some threshhold into a mind-killing, mundane amount of repetition.
This is a great piece, appreciate you sharing it! I have mixed feelings on the situation. I 100% understand what the author is saying, and while I definitely haven't experienced it as much as the...
This is a great piece, appreciate you sharing it!
I have mixed feelings on the situation. I 100% understand what the author is saying, and while I definitely haven't experienced it as much as the author clearly has, I know the bliss of day-long flow states that they're describing. I've experienced it with both coding, 3D animation/modeling, and hands-on maker projects.
AI has changed it, and I both lament and celebrate it's arrival. On the one hand I absolutely agree, the rhythm has changed. Before, your hands couldn't possibly type as fast as you could think and so you have to code and create at a moderate but consistent pace. It inherently provides room to think further ahead, to second guess decisions and think of new alternatives, and spot potential bugs along the way. With AI you can now create at a speed that would have been inconceivable before. The code appearing now outpaces your thoughts. You are constantly making progress, constantly moving forward, constantly thinking of features to add. Syntax errors are almost nonexistant and if they appear they are fixed with a proverbial snap of the fingers. Runtime errors are almost as easily diagnosed and fixed (this of course depends on how complex the overall system is).
So all that's left is the subtle bugs, the systemic errors, the mistakes that are often small and easy to overlook but fundamentally change how a program needs to operate. And it's these things in particular that worry me, because no AI can anticipate things about which it has no context. Moving slower could very likely leave room to discover or realize oversights like this, but AI moves so fast it leaves no room to consider.
I worry that the new programmer flow state in the age of AI is just an even worse version of "move fast and break things". I hope we figure something better than that, though I don't know what that is.
I feel like people are either all in on using agents or rejecting them entirely. Why don’t more people use them in moderation? Spend half of the day fielding interrupts, and half focusing on the...
I feel like people are either all in on using agents or rejecting them entirely. Why don’t more people use them in moderation? Spend half of the day fielding interrupts, and half focusing on the most important work? Maybe some people are afraid to ever take it slow because they’re judged by their LLM usage stats.
Before agentic coding I heard about AI firms where you were judged by your GPU runtime every month. People would spin up duplicate jobs just to pump up their numbers. I would do the same thing to burn tokens if I was in a toxic workspace that judged developers by their LLM use.
I don't have anything to back this up, but my impression is that there are some ways in which each of the two approaches doesn't help you improve at the other or even trains you detrimentally for...
I don't have anything to back this up, but my impression is that there are some ways in which each of the two approaches doesn't help you improve at the other or even trains you detrimentally for the respective other.
The skill set needed to leverage agents optimally is akin to that of a technical manager.
This is different from the specialized skills of someone focusing on deep work on individual contributions.
Also:
Focusing a lot on deep work in your own style can make you too opinionated (this is my biggest problem with using LLMs right now) on some properties of a solution, which will make you frustrated with the solution of an LLM and also slow the LLM down
Focusing on using agents can atrophy your ability to bang your head against the wall until you find your own solution.
Focusing on using agents will also just mean you have less practice relevant to the minutiae of working on stuff yourself.
I'm in the middle. I love the AI workflow for some coding tasks and don't for others. I'm lucky to be at a job where we aren't being judged in any way by our usage of AI so I get to find that...
I'm in the middle. I love the AI workflow for some coding tasks and don't for others. I'm lucky to be at a job where we aren't being judged in any way by our usage of AI so I get to find that middle ground for myself. I'm very aware of the skill degredation that comes along with reliance on AI and I'm also wary of the mediocrity and hallucinations. The more I know about a topic the more I see how limited it is. But there are times when I save hours on menial coding tasks and that is an undeniable boost and feels great!
I'm not certain but I want to say more people do use it in moderation, but strong opinions rise to the top. How many times does nuance and "it's complicated" hit the top of the algorithm? Perhaps also an element that there's such strong feelings around the tech that even what counts as moderation is up for debate? I vibe more with the "hate it" crowd but may not be considered a moderate user by those who want zero AI usage seeing as I use it daily at work. And I'm sure I seem almost Luddite for those who have gone all in on agentic usage.
I guess at the end of the day this is a tech with scary potential and that makes it easy for opinionated people to grab a lot of attention. A lot of us are looking for answers and by golly there are a lot of answers out there haha
I'm going to leave the AI stuff aside, because I've talked to death about it, and I'm sure we all have, and I'm kind of tired of thinking and hearing about it. What's more interesting to me is...
I'm going to leave the AI stuff aside, because I've talked to death about it, and I'm sure we all have, and I'm kind of tired of thinking and hearing about it.
What's more interesting to me is that to me, it doesn't seem like the author of this article's problem is actually AI related at all. His problem seems like his entire life is one dimensional.
I don't know if he was exaggerating for effect or what, but if the only thing you enjoy doing, and the only thing driving you is truly programming and listening to Phish, you're cheating yourslef out of life. Like... Badly.
Maybe the author is not neurotypical and doing other things is uncomfortable for him or something, but man... that was one of the most depressing descriptions of a life I've ever heard, and that's even before I got to the part where he had a problem.
I definitely understand being passionate about one thing (or two things I guess), but not when it's at the expense of everything else. Falling in love, caring for a pet, seeing other cultures, playing a sport, developing a new skill, tending a garden, and so on and so on. Life has an infinite amount of things to do, most people won't be into most of them, but that's okay. If you're not at least exposing yourself to some of them though, especially when you have the means to do so?
Man, that's just such a shame to just decide at 15 that you're only ever going to program and listen to Phish for the rest of your life.
Also, one half of his life has already been ruined by AI coding agents. What happens if Trey turns out to secretly be a Nazi or child molester or something? Is your entire life just effectively over?
I interpreted this piece as being about his work, so I didn't assume that he eschews everything else life has to offer. I only know him from this one article so I wouldn't assume that he doesn't...
I interpreted this piece as being about his work, so I didn't assume that he eschews everything else life has to offer. I only know him from this one article so I wouldn't assume that he doesn't have love in his life, or doesn't care for creatures/plants, etc. That said, if his life is completely filled up by programming while listening to music and by traveling to watch live music-- I don't necessarily see it as sad or one-dimensional either, since I see both of those activities as multifaceted and potentially filled with a lot of different experiences already.
This reminds me of some of Linus Torvald's writings from a decade or two ago. I remember reading his opinion that he had switched from writing code for the Linux kernel to basically managing the...
I don’t really write code anymore. The main thing now is managing agents. I open a session, ask a question, redirect, switch to a different one, check on a merge, review what came back, send it back for changes, switch again. The day is a queue. Things finish at different times and require different responses, and the responses are short and the contexts are constantly different.
This reminds me of some of Linus Torvald's writings from a decade or two ago. I remember reading his opinion that he had switched from writing code for the Linux kernel to basically managing the git merges full time. It sucks that this person is going through it. I felt it in my previous job when I went from being an individual contributor to a manager. I absolutely dreaded being in charge of other people's output, and not being held accountable for my own any longer. Some people excel at that type of work, and others aren't cut out for it.
I never learned to code, myself. So I don't feel qualified to comment if the invention of LLMs is actually the death knell for programming by hand or not. Past experience of similar situations have me thinking that it likely isn't the case, though: because typing didn't eradicate writing by hand, and self driving cars didn't save us all from driving for ourselves. But I see the other side, too: the assembly line has made it so no one assembles products fully by hand any longer.
To your point on the assembly line replacing hand assembly, handmade objects are a niche that some people do value, and are seen as a luxury for certain industries. I wonder if something similar...
To your point on the assembly line replacing hand assembly, handmade objects are a niche that some people do value, and are seen as a luxury for certain industries. I wonder if something similar could happen with software? I don't imagine there being "luxury software" in the future but software built by hand with the care of a human.
Did it not? I’m pressed to think of a current industry that regularly produces handwritten documents. The only example I can think of is schooling, but that’s because handwriting is better for...
typing didn't eradicate writing by hand
Did it not? I’m pressed to think of a current industry that regularly produces handwritten documents. The only example I can think of is schooling, but that’s because handwriting is better for memory. I can surely believe people will continue to code by hand in school to learn the basics if they are studying computer science.
self driving cars didn't save us all from driving for ourselves
Self-driving cars clearly aren’t ready yet, but the industry still has room to grow. Waymo works pretty well in San Francisco. I do not think humans will drive their own cars in 50 years, but these changes take time.
My thoughts came from the context of the blog post and the comment I responded to; both are about work. I believe that the industrial capacity for an activity is a good proxy for whether or not...
My thoughts came from the context of the blog post and the comment I responded to; both are about work. I believe that the industrial capacity for an activity is a good proxy for whether or not that activity is a viable career.
The author of the post is not prevented from listening to Phish and coding by hand; I still do most of my hobby programming by hand. They’re only noticing that, if they want to continue to be paid for their work, that programming by hand is less viable. I think that’s analogous to handwriting.
I don't see how. Typing and handwriting are mechanical, with a difference in how your mind interfaces with the scribing itself, but only speed distinguishes the actual composition process between...
I think that’s analogous to handwriting.
I don't see how. Typing and handwriting are mechanical, with a difference in how your mind interfaces with the scribing itself, but only speed distinguishes the actual composition process between the two, and plenty of authors switch between handwriting and typing just for ergonomic sake, or draft in hand and proof in type, etc. Programming, on the other hand, is not the same thing as managing nonhuman programmers. At all. They're entirely different skillsets with different impacts on the world. I challenged your point because it seemed specious, but now it seems you're just pulling a "coal miners: learn to code!"
The analogy I had in mind was that, in both cases, it appears that a new technology makes a previous skill much less valuable. (Though programming by hand might make a comeback after the bubble...
The analogy I had in mind was that, in both cases, it appears that a new technology makes a previous skill much less valuable. (Though programming by hand might make a comeback after the bubble pops.)
They're entirely different skillsets with different impacts on the world.
I agree that the analogy falls apart here.
I challenged your point because it seemed specious, but now it seems you're just pulling a "coal miners: learn to code!"
I claimed that handwriting is no longer around in a commercially relevant way. When you say I’m pulling a “coal miners: learn to code!”, is it because you’re reading “coders: learn to manage agents!”? I agree that they are different sets of skills, and I do think one needs to find work that suits their skills. I’m just quite skeptical about the future of programming as a career.
I personally think I’m a competent programmer, but I left software engineering entirely (for the legal field) in response to the rise of LLMs, since I hate the scope creep agents present.
Ah, fair enough. That does shift how I interpret your past comments, and as such I don't have much to say in disagreement. Thank you for clarifying for me.
I’m just quite skeptical about the future of programming as a career.
Ah, fair enough. That does shift how I interpret your past comments, and as such I don't have much to say in disagreement. Thank you for clarifying for me.
Was very moved by this personal essay about generative AI degrading flow state.
It's relatable for me even though I'm not a professional programmer. It's not necessarily AI specifically (although that's definitely a significant factor for me), but also things like rapid context-switching and algorithm-controlled feeds (which I both consume content from and produce content for).
I've been a professional visual artist most of my adult life, but I've been struggling with getting into a flow state with that recently. I'm currently learning game development and I really enjoy coding (even if I'm not good at it and make a lot of spaghetti!), so it's been easier for me to get into a flow state with that at the moment.
Excerpts below:
The discussion about this on Hacker News may also be of interest, since there are others sharing differing perspectives. For some, coding with AI has improved flow state since they prefer working rapidly with ideas / design / architecture.
This reads like working on an assembly line or in a warehouse, but instead it's on a pc.
My worst job was order picking in my teens. Every menial task is in a queue waiting to be picked up. No meaningful variation. You work through it until you clock out. Next day however, another queue piled up.
There's little more soul crushing than monotony like that.
I don't think it's much like working in a warehouse. It might be like a support job (responding to trouble tickets) except that you never get abuse from a coding agent, you can rely on some baseline intelligence even if it does screw up sometimes, and you're the one giving orders. (Or suggestions, which is how I usually phrase it.)
The multitasking is because LLM's are often too slow to wait around for; if they were faster then there wouldn't be much point switching tasks.
Since I'm just coding for fun, I multitask with something else (like checking Tildes or doing something else entirely) rather than trying to multitask programming.
Oh no I'm aware right now it's nothing like working in a warehouse, just that description is eerily similar. Nevertheless, a future where all a programmer does is verify code written by AI isn't too far off menial work.
I believe you!
Your comment reminded me of this personal essay by Aella where she talks about how she started on this job feeling so excited and eager to learn and improve, but it just turned out to be soul-crushing tedium: When I was 19, I spent a year assembling electrical relays on a factory floor for $10/hr.
Excerpt:
This is a great piece, appreciate you sharing it!
I have mixed feelings on the situation. I 100% understand what the author is saying, and while I definitely haven't experienced it as much as the author clearly has, I know the bliss of day-long flow states that they're describing. I've experienced it with both coding, 3D animation/modeling, and hands-on maker projects.
AI has changed it, and I both lament and celebrate it's arrival. On the one hand I absolutely agree, the rhythm has changed. Before, your hands couldn't possibly type as fast as you could think and so you have to code and create at a moderate but consistent pace. It inherently provides room to think further ahead, to second guess decisions and think of new alternatives, and spot potential bugs along the way. With AI you can now create at a speed that would have been inconceivable before. The code appearing now outpaces your thoughts. You are constantly making progress, constantly moving forward, constantly thinking of features to add. Syntax errors are almost nonexistant and if they appear they are fixed with a proverbial snap of the fingers. Runtime errors are almost as easily diagnosed and fixed (this of course depends on how complex the overall system is).
So all that's left is the subtle bugs, the systemic errors, the mistakes that are often small and easy to overlook but fundamentally change how a program needs to operate. And it's these things in particular that worry me, because no AI can anticipate things about which it has no context. Moving slower could very likely leave room to discover or realize oversights like this, but AI moves so fast it leaves no room to consider.
I worry that the new programmer flow state in the age of AI is just an even worse version of "move fast and break things". I hope we figure something better than that, though I don't know what that is.
I feel like people are either all in on using agents or rejecting them entirely. Why don’t more people use them in moderation? Spend half of the day fielding interrupts, and half focusing on the most important work? Maybe some people are afraid to ever take it slow because they’re judged by their LLM usage stats.
Before agentic coding I heard about AI firms where you were judged by your GPU runtime every month. People would spin up duplicate jobs just to pump up their numbers. I would do the same thing to burn tokens if I was in a toxic workspace that judged developers by their LLM use.
I don't have anything to back this up, but my impression is that there are some ways in which each of the two approaches doesn't help you improve at the other or even trains you detrimentally for the respective other.
Also:
I'm in the middle. I love the AI workflow for some coding tasks and don't for others. I'm lucky to be at a job where we aren't being judged in any way by our usage of AI so I get to find that middle ground for myself. I'm very aware of the skill degredation that comes along with reliance on AI and I'm also wary of the mediocrity and hallucinations. The more I know about a topic the more I see how limited it is. But there are times when I save hours on menial coding tasks and that is an undeniable boost and feels great!
I'm not certain but I want to say more people do use it in moderation, but strong opinions rise to the top. How many times does nuance and "it's complicated" hit the top of the algorithm? Perhaps also an element that there's such strong feelings around the tech that even what counts as moderation is up for debate? I vibe more with the "hate it" crowd but may not be considered a moderate user by those who want zero AI usage seeing as I use it daily at work. And I'm sure I seem almost Luddite for those who have gone all in on agentic usage.
I guess at the end of the day this is a tech with scary potential and that makes it easy for opinionated people to grab a lot of attention. A lot of us are looking for answers and by golly there are a lot of answers out there haha
I'm going to leave the AI stuff aside, because I've talked to death about it, and I'm sure we all have, and I'm kind of tired of thinking and hearing about it.
What's more interesting to me is that to me, it doesn't seem like the author of this article's problem is actually AI related at all. His problem seems like his entire life is one dimensional.
I don't know if he was exaggerating for effect or what, but if the only thing you enjoy doing, and the only thing driving you is truly programming and listening to Phish, you're cheating yourslef out of life. Like... Badly.
Maybe the author is not neurotypical and doing other things is uncomfortable for him or something, but man... that was one of the most depressing descriptions of a life I've ever heard, and that's even before I got to the part where he had a problem.
I definitely understand being passionate about one thing (or two things I guess), but not when it's at the expense of everything else. Falling in love, caring for a pet, seeing other cultures, playing a sport, developing a new skill, tending a garden, and so on and so on. Life has an infinite amount of things to do, most people won't be into most of them, but that's okay. If you're not at least exposing yourself to some of them though, especially when you have the means to do so?
Man, that's just such a shame to just decide at 15 that you're only ever going to program and listen to Phish for the rest of your life.
Also, one half of his life has already been ruined by AI coding agents. What happens if Trey turns out to secretly be a Nazi or child molester or something? Is your entire life just effectively over?
I interpreted this piece as being about his work, so I didn't assume that he eschews everything else life has to offer. I only know him from this one article so I wouldn't assume that he doesn't have love in his life, or doesn't care for creatures/plants, etc. That said, if his life is completely filled up by programming while listening to music and by traveling to watch live music-- I don't necessarily see it as sad or one-dimensional either, since I see both of those activities as multifaceted and potentially filled with a lot of different experiences already.
This reminds me of some of Linus Torvald's writings from a decade or two ago. I remember reading his opinion that he had switched from writing code for the Linux kernel to basically managing the git merges full time. It sucks that this person is going through it. I felt it in my previous job when I went from being an individual contributor to a manager. I absolutely dreaded being in charge of other people's output, and not being held accountable for my own any longer. Some people excel at that type of work, and others aren't cut out for it.
I never learned to code, myself. So I don't feel qualified to comment if the invention of LLMs is actually the death knell for programming by hand or not. Past experience of similar situations have me thinking that it likely isn't the case, though: because typing didn't eradicate writing by hand, and self driving cars didn't save us all from driving for ourselves. But I see the other side, too: the assembly line has made it so no one assembles products fully by hand any longer.
To your point on the assembly line replacing hand assembly, handmade objects are a niche that some people do value, and are seen as a luxury for certain industries. I wonder if something similar could happen with software? I don't imagine there being "luxury software" in the future but software built by hand with the care of a human.
Did it not? I’m pressed to think of a current industry that regularly produces handwritten documents. The only example I can think of is schooling, but that’s because handwriting is better for memory. I can surely believe people will continue to code by hand in school to learn the basics if they are studying computer science.
Self-driving cars clearly aren’t ready yet, but the industry still has room to grow. Waymo works pretty well in San Francisco. I do not think humans will drive their own cars in 50 years, but these changes take time.
Why do you conflate the existence of a phenomenon with the industrial capacity for the phenomenon?
My thoughts came from the context of the blog post and the comment I responded to; both are about work. I believe that the industrial capacity for an activity is a good proxy for whether or not that activity is a viable career.
The author of the post is not prevented from listening to Phish and coding by hand; I still do most of my hobby programming by hand. They’re only noticing that, if they want to continue to be paid for their work, that programming by hand is less viable. I think that’s analogous to handwriting.
I don't see how. Typing and handwriting are mechanical, with a difference in how your mind interfaces with the scribing itself, but only speed distinguishes the actual composition process between the two, and plenty of authors switch between handwriting and typing just for ergonomic sake, or draft in hand and proof in type, etc. Programming, on the other hand, is not the same thing as managing nonhuman programmers. At all. They're entirely different skillsets with different impacts on the world. I challenged your point because it seemed specious, but now it seems you're just pulling a "coal miners: learn to code!"
The analogy I had in mind was that, in both cases, it appears that a new technology makes a previous skill much less valuable. (Though programming by hand might make a comeback after the bubble pops.)
I agree that the analogy falls apart here.
I claimed that handwriting is no longer around in a commercially relevant way. When you say I’m pulling a “coal miners: learn to code!”, is it because you’re reading “coders: learn to manage agents!”? I agree that they are different sets of skills, and I do think one needs to find work that suits their skills. I’m just quite skeptical about the future of programming as a career.
I personally think I’m a competent programmer, but I left software engineering entirely (for the legal field) in response to the rise of LLMs, since I hate the scope creep agents present.
Ah, fair enough. That does shift how I interpret your past comments, and as such I don't have much to say in disagreement. Thank you for clarifying for me.