The author makes two assumptions that I personally don’t agree with when using AI tooling:
that the projects they made no longer see any use, and
that the only value is from marketing them, startup style.
I have the opposite problem: I use my vibe coded utilities on a daily basis, and I have no intention of ever sharing them with the outside world. As of now, at least.
yup. I’ve made several very cool things in the last month which probably are worth sharing with a certain pool of people, but I am not a marketer and don’t like shilling/self promotion, so they’ll likely stay incredibly niche and helpful to me specifically.
but I also don’t have to feel too bad about it, any engineer who cares as much as me and is willing to pay for claude max for a month could certainly reproduce nearly the same things, but to their own taste.
I have no intention of ever sharing them with the outside world
Yes! This is a sad part for me.
If I shared some of the tools I vibecoded (despite being very useful to me) on a wider basis, I'd be uncomfortable because I'd be giving others a tool that doesn't reflect me and that I don't understand.
Net result is that these stay private, even though some of them have some merit... The effort to polish them up to the state I'd feel happy sharing them is too high though.
I've actually fixed many problems, or got things working in the way I want, by stumbing on someone's dotfiles which either contain the exact thing I need, or point me in the right direction. That's never required such things to be organised, readable, documented, etc. it's just been blobs of code surfaced in Google results for particular combinations of keywords. The more that's out there, the better IMHO.
I share my dotfiles and scripts (some vibe-coded), but mostly in a "thrown over the wall" way, with no implication that it will be useful, correct, or that I will see (let alone respond to) any contributions. At the very least, having copies makes it harder to lose.
On that last point, this technology is horrific for attention. It's a thermonuclear ADHD amplifier and I have seen the same effect in every single one of my adult friends. Folk running 3 screens simultaneously working on totally unrelated "projects" they have little hope of maintaining, and such little commitment to the outcome that the time is obviously wasted.
This rings true to me. My ADHD is sometimes fed well by AI, being able to stimulate on 3 or 4 projects at once, cycling through guiding the agent along and doing review on each one.
But it's also exhausting, and I haven't been "in the zone" nearly as much as I used to. Also, at the end of the day my understanding of the projects and the code I shipped just doesn't feel as "solid" as it used to (because it isn't).
I think it's probably akin to moving from a Senior Engineering role into a management or similar role, where you have to give up intimately knowing the details and just trust that others have done a good enough job.
I feel similarly to you and also find it exhausting. I also feel that, even though I’m outputting more completed work than ever, I have a reluctance to share the work because of the reduced deep understanding.
When I made my most recent job change, I moved from a management role back into a senior engineering role. I've been writing and reviewing code through that time and I can still do it, but internally? I came back feeling less confident in my skills and attention to detail.
That's not entirely because of the management switch: I moved jobs because I was getting very burnt out, so those things were already slipping on their own.
But after a while being encouraged to use coding agents at work I'm finding that for me, LLM-assisted coding makes that worse. I'm using it less, because I don't necessarily trust myself to review it fully.
The flip-side: I'm more inclined to start side projects than I have been in the past, because I want to hand-code them and (re)build my skills.
i'm finding the exact opposite effect, with LLMs on my ADHD
i don't exactly expect this to be a universal experience. but i find when i don't want to engage with my main project, i can often fire off claude to go do something, and then when i come back, i have an already started problem to chew on. reviewing it, making changes to it, using it to do the next thing, etc.
i've been building a game engine since ~december. this is probably the longest i've stuck to a single project without external pressure. and i have not needed to take my ritalin as often
(i've actually found that LLMs+ritalin can be a bit "too much", though i'm not exactly sure what it's too much of)
I mean… of course it’s easy to stay on a project when an LLM is doing all of the coding. I find that aspect of LLMs too alluring and it’s really wrecking students abilities to write academically, ie think critically.
On the flip side, when I struggle with writing from scratch (getting started just feels mentally impossible), editing is also what helps me get in the zone so I fully agree with your point.
Yes, agreed on both points. LLMs can definitely unblock when stuck in a rut, but I worry that if I actually come to rely on this then I will lose the ability to power through under my own steam.
I already at some point got too lazy to search for things manually at some point (not helped by the deteriorating quality of traditional search engine results), finding my fingers reaching for 'cla[ude...]' or 'cha[tgpt...]' instead of search..
I am trying to pull out of that since I can feel myself losing the ...rigour(?) I once had.
I found this very convincing personally, but it was interesting how many comments in the Hacker News thread reported the opposite from an ADHD perspective. Quoting three:
"... for me (also ADHD) it's kind of the opposite. I'm finishing side projects for the first time ever because I can actually get them working before I get bored of them"
"As someone with ADHD I feel like AI is a salve for my mind. I used to listen to intense EDM while working. Now I sit in silence and talk to my agents. I maintain inbox zero. I absorb and comment across all relevant projects, even outside my team. I literally feel like I have a support team for the first time."
"For those of us prone to hyperfocus, working with AI can provide the kinds of stimulation we crave. I can hardly remember a time when I've felt more engaged with my work, more productive, and more badass."
Hi Simon, thanks for the syndication :) Now I wish I'd spent an extra half hour on editing..
I think re: the ADHD thing, there may be some element of causing affront in suggesting work may have no value provoking a natural response. It's easy to feel supercharged and productive working with AI tools, I personally consider that an illusion. Relating work to a substantial end goal beyond code complete was a major point I quite poorly muddled. Everyone is obviously getting a huge kick from using the tools, I suppose the question of substance is a personal one.
On my blog I said "I find this post by David Wilson very relatable" and I really do - I'm in the same kind of position as you, I've been spinning off new projects at an absurd rate and just recently started to review them and realized that quite a lot of them really are just distractions from what I want to get done.
I have a collaborative editable whiteboard Datasette plugin on my laptop right now with an API so Claude Code can collaborate with me (and I got it to draw a pelican) - it's really cool! Nobody needs that, and realistically there's no way I'm going to knock it into a releasable shape.
i could be wrong, but from what i can tell, most of your projects are fairly open ended (datasette and llm look like the core works, and you can go many directions to improve them. and your blog is a blog)
my hunch is that that's a major part of the split between "helps stay on track" and "infinite well of distractions" is how concrete and narrow your goals are
even with a concrete goal, i find it's easy to get distracted, want to restart, etc.
but when you have that many options of things to do next (and one of them is "do interesting things to post about"), i'd guess it's much more prone to being distracting. but, distracting from what?
i'll see if i hit similar issues once my project gets to the point where the design space opens up, i suppose
Yeah, Datasette is built around plugins, and the joy of plugins is that almost anything I want to play around with can be justified as a Datasette plugin.
For me, it's definitly an adhd multiplier, bit i'd say: still in a good sense.
i'm using ai mostly to figure out what i actually want on a high-level viewpoint, then i design an mental architecture for that project and use llms to write the code.
this means i already know fairly well where i wanna go, and my prompts often already include specific code snippets i want to use for implementing things.
for me, many small prompts have shown way better results that less big prompts, as i'm staying way more steerable.
i'm also only focusing on two projects right now, and not simultaneously, so the focus is stronger than without llms, which is kinda surprising to me.
I've personally found with my flavor of ADHD that LLMs are horrifically distracting and detrimental to focusing and accomplishing tasks. It's like having another person in the room who never stops talking about mostly unrelated stuff.
I'm struggling with coming up with new ideas to pursue. I think with the friction of researching and implementing projects in the past a lot of new ideas surfaced. Many of those I also deemed not worth it to even bother with. Now I can validate them with very low effort with an LLM and because I'm not solving as many problems myself with deep focus and problem solving, nothing comes out of it except a POC for an idea that might not even be worth much (not just in monetary sense).
EDIT: For whatever it's worth I also have an ADHD diagnosis and doing side projects was a great source of dopamine for me and as a result I feel like I'm overall unhappier.
briankung | 20 hours ago
The author makes two assumptions that I personally don’t agree with when using AI tooling:
I have the opposite problem: I use my vibe coded utilities on a daily basis, and I have no intention of ever sharing them with the outside world. As of now, at least.
lilac | 17 hours ago
yup. I’ve made several very cool things in the last month which probably are worth sharing with a certain pool of people, but I am not a marketer and don’t like shilling/self promotion, so they’ll likely stay incredibly niche and helpful to me specifically.
but I also don’t have to feel too bad about it, any engineer who cares as much as me and is willing to pay for claude max for a month could certainly reproduce nearly the same things, but to their own taste.
reivilibre | 10 hours ago
Yes! This is a sad part for me. If I shared some of the tools I vibecoded (despite being very useful to me) on a wider basis, I'd be uncomfortable because I'd be giving others a tool that doesn't reflect me and that I don't understand. Net result is that these stay private, even though some of them have some merit... The effort to polish them up to the state I'd feel happy sharing them is too high though.
andyferris | 5 hours ago
Maybe these are just the modern equivalent of scripts and dotfiles we have floating around our computers?
The cost of producing them was low, the level of customisation was high.
Sharing simply isn’t helping anyone else - feeling odd/sad/guilty about not sharing is equally not helping you.
chriswarbo | 3 hours ago
I've actually fixed many problems, or got things working in the way I want, by stumbing on someone's dotfiles which either contain the exact thing I need, or point me in the right direction. That's never required such things to be organised, readable, documented, etc. it's just been blobs of code surfaced in Google results for particular combinations of keywords. The more that's out there, the better IMHO.
I share my dotfiles and scripts (some vibe-coded), but mostly in a "thrown over the wall" way, with no implication that it will be useful, correct, or that I will see (let alone respond to) any contributions. At the very least, having copies makes it harder to lose.
schell | 21 hours ago
This rings true to me. My ADHD is sometimes fed well by AI, being able to stimulate on 3 or 4 projects at once, cycling through guiding the agent along and doing review on each one.
But it's also exhausting, and I haven't been "in the zone" nearly as much as I used to. Also, at the end of the day my understanding of the projects and the code I shipped just doesn't feel as "solid" as it used to (because it isn't).
I think it's probably akin to moving from a Senior Engineering role into a management or similar role, where you have to give up intimately knowing the details and just trust that others have done a good enough job.
anex9d | 16 hours ago
I feel similarly to you and also find it exhausting. I also feel that, even though I’m outputting more completed work than ever, I have a reluctance to share the work because of the reduced deep understanding.
lilac | 12 hours ago
incredibly relatable. I’m willing to subject myself to the unknowns but less so others.
owent | 3 hours ago
When I made my most recent job change, I moved from a management role back into a senior engineering role. I've been writing and reviewing code through that time and I can still do it, but internally? I came back feeling less confident in my skills and attention to detail.
That's not entirely because of the management switch: I moved jobs because I was getting very burnt out, so those things were already slipping on their own.
But after a while being encouraged to use coding agents at work I'm finding that for me, LLM-assisted coding makes that worse. I'm using it less, because I don't necessarily trust myself to review it fully.
The flip-side: I'm more inclined to start side projects than I have been in the past, because I want to hand-code them and (re)build my skills.
hc | 19 hours ago
i'm finding the exact opposite effect, with LLMs on my ADHD
i don't exactly expect this to be a universal experience. but i find when i don't want to engage with my main project, i can often fire off claude to go do something, and then when i come back, i have an already started problem to chew on. reviewing it, making changes to it, using it to do the next thing, etc.
i've been building a game engine since ~december. this is probably the longest i've stuck to a single project without external pressure. and i have not needed to take my ritalin as often (i've actually found that LLMs+ritalin can be a bit "too much", though i'm not exactly sure what it's too much of)
oceanhaiyang | 17 hours ago
I mean… of course it’s easy to stay on a project when an LLM is doing all of the coding. I find that aspect of LLMs too alluring and it’s really wrecking students abilities to write academically, ie think critically.
On the flip side, when I struggle with writing from scratch (getting started just feels mentally impossible), editing is also what helps me get in the zone so I fully agree with your point.
reivilibre | 10 hours ago
Yes, agreed on both points. LLMs can definitely unblock when stuck in a rut, but I worry that if I actually come to rely on this then I will lose the ability to power through under my own steam.
I already at some point got too lazy to search for things manually at some point (not helped by the deteriorating quality of traditional search engine results), finding my fingers reaching for 'cla[ude...]' or 'cha[tgpt...]' instead of search.. I am trying to pull out of that since I can feel myself losing the ...rigour(?) I once had.
simonw | 18 hours ago
I found this very convincing personally, but it was interesting how many comments in the Hacker News thread reported the opposite from an ADHD perspective. Quoting three:
dw | 17 hours ago
Hi Simon, thanks for the syndication :) Now I wish I'd spent an extra half hour on editing..
I think re: the ADHD thing, there may be some element of causing affront in suggesting work may have no value provoking a natural response. It's easy to feel supercharged and productive working with AI tools, I personally consider that an illusion. Relating work to a substantial end goal beyond code complete was a major point I quite poorly muddled. Everyone is obviously getting a huge kick from using the tools, I suppose the question of substance is a personal one.
simonw | 15 hours ago
On my blog I said "I find this post by David Wilson very relatable" and I really do - I'm in the same kind of position as you, I've been spinning off new projects at an absurd rate and just recently started to review them and realized that quite a lot of them really are just distractions from what I want to get done.
I have a collaborative editable whiteboard Datasette plugin on my laptop right now with an API so Claude Code can collaborate with me (and I got it to draw a pelican) - it's really cool! Nobody needs that, and realistically there's no way I'm going to knock it into a releasable shape.
hc | 14 hours ago
i could be wrong, but from what i can tell, most of your projects are fairly open ended (datasette and llm look like the core works, and you can go many directions to improve them. and your blog is a blog)
my hunch is that that's a major part of the split between "helps stay on track" and "infinite well of distractions" is how concrete and narrow your goals are
even with a concrete goal, i find it's easy to get distracted, want to restart, etc. but when you have that many options of things to do next (and one of them is "do interesting things to post about"), i'd guess it's much more prone to being distracting. but, distracting from what?
i'll see if i hit similar issues once my project gets to the point where the design space opens up, i suppose
simonw | 12 hours ago
Yeah, Datasette is built around plugins, and the joy of plugins is that almost anything I want to play around with can be justified as a Datasette plugin.
xq | 9 hours ago
For me, it's definitly an adhd multiplier, bit i'd say: still in a good sense.
i'm using ai mostly to figure out what i actually want on a high-level viewpoint, then i design an mental architecture for that project and use llms to write the code.
this means i already know fairly well where i wanna go, and my prompts often already include specific code snippets i want to use for implementing things.
for me, many small prompts have shown way better results that less big prompts, as i'm staying way more steerable.
i'm also only focusing on two projects right now, and not simultaneously, so the focus is stronger than without llms, which is kinda surprising to me.
stargirl | 3 hours ago
I've personally found with my flavor of ADHD that LLMs are horrifically distracting and detrimental to focusing and accomplishing tasks. It's like having another person in the room who never stops talking about mostly unrelated stuff.
reezer | 8 hours ago
All of the author's friends are diagnosed with ADHD?!
granra | 3 hours ago
I'm struggling with coming up with new ideas to pursue. I think with the friction of researching and implementing projects in the past a lot of new ideas surfaced. Many of those I also deemed not worth it to even bother with. Now I can validate them with very low effort with an LLM and because I'm not solving as many problems myself with deep focus and problem solving, nothing comes out of it except a POC for an idea that might not even be worth much (not just in monetary sense).
EDIT: For whatever it's worth I also have an ADHD diagnosis and doing side projects was a great source of dopamine for me and as a result I feel like I'm overall unhappier.