At the very least, I do encourage people who are working on personal projects for explicitly learning and studying purposes to purposefully not use AI. Of course, if the purpose of the project is to learn AI, then that rule is allowed to be broken.
Anecdotal example but I know plenty of students (Including my own brother) who graduated recently who did it almost entirely with AI (unfortunately at the request of their own professors) who can't answer the most basic of questions that I would have expected of sometimes even first year CS students in the days before AI was a thing, which is why I encourage them to build projects without it.
Personally, I'm starting to think that in university and education systems, all projects should now be done in class and in computer labs/monitored study halls because we can't trust anyone to not use AI anymore.
I've been doing this on my latest personal project. What I've realised I missed the most is feeling in the zone or in a "flow" state as some call it. It really feels amazing. Coding agents take that away from you. I really think we need some coding agent implementation which takes less agency away from programmers and helps people be in the flow even more.
For my project I use Gemini/Google's AI mode to great effect. Quick answers to be reminded about syntax, and to get help when stuck. It actually enhances the flow state. Plus it's all free.
I don't think it has to do with sound. Some sounds can disrupt it, of course. I can work well in a noisy cafe as long as I'm not listening to any particular conversation. Although oftentimes, if I'm alone, I talk to myself out loud while I'm coding ("no, not that, it should go here... where's that header? stupid typo...")
We used to call it "seeing through the code". The moment when you're no longer looking at text on a page, you're seeing what it does as you scroll through it.
What you need most, to get to that place, is to simply spend a lot of time writing and reading code until it becomes second nature. Something like 10,000 hours.
Once you've done that, reading an LLM's code is trivial. But letting it write most of the code, without reviewing it line by line, causes rapid brain atrophy. Because to really be in the zone, you need to keep everything relevant in your head. Which means it has to be in there in the first place.
Eh, experiences will vary. For me, my zone is conducting things, not typing. Firing something off in twelve tabs and coming back later is precisely how my brain works.
> Firing something off in twelve tabs and coming back later
As someone who gets "in the zone" like the person you're replying to, I just don't get how this could possibly be the same thing. "Coming back later" doesn't make any sense when you're in that state. Stopping and waiting is a killer. You're just constantly doing things without necessarily consciously thinking about it.
Do you mean you're switching continually between the twelve sessions? Or literally "coming back" in a different sort of operating mode?
The closest thing to it for me using AI tools is specifically not running multiple "agents" but making targeted changes over and over and over but with a 10x+ increase in typing-vs-change ratio. (This also rules out things like Opus let alone Fable as they're just too slow and do too much at once.)
I got out of tech about 5 years ago, before AI really took off. I still do a lot of personal projects and now never use AI. I tried it a bit (only free plans) but noticed how much it caused me to no longer be able to think through a solution myself. I thought I was learning a lot from it but for some reason whatever I learned from it didn't stick. So now I'm no longer using it at all. I do realize that this is very specific to my situation of course.
There was a well known phenomenon people called ‘tutorial hell’ where when learning to program you could understand all the code from tutorials, but when it came time to actually create your own solutions for stuff you got stuck. I think AI is basically enabling eternal tutorial hell, you never actually need to break out of it anymore.
Difference between taking your own notes and skimming someone else's.
I have found in the last year I "get more done" but I learn less. Whats even worse is if I burn though my token quota I am less likely(and less able) to "switch to manual". LLMs make an ocean of decisions and assumptions very quickly and if you don't slow it down very intentionally (and lose the promise of the AI providers are pitching) you will quickly not have a good idea of how systems work.
It’s funny, I’m the exact opposite. I’ve tackled projects at scales I could never have done before, and learned more than I would have.
My actual coding ability is probably lower, but I’ve learned so much about music theory and production, CAD, and statistics. All because my hobby projects can go so much further than before.
Isn't there a risk that this learning is too superficial to actually stay around and impact your thinking in the long run?
How I experience it at least, there is no free lunch. Something that gives you knowledge and experience will almost by definition be painful and uncomfortable.
We use AI because it is faster, but that speed has a consequence.
I’ve never been a cram for the test kind of person. I do best when using knowledge in an ongoing basis. Regardless of AI, things I studied intensely a decade ago but never use are by and large gone, whereas things I use frequently stay and deepen.
I don’t expect it to be any different with AI, but YMMV.
I really do not get all of y'all claiming this causes you to "not be able to think".
I worked on six nines active active payments infra that handled billions in daily transaction volume. I've built laser projection systems to play interactive games on skyscrapers. That ability is not atrophying.
I'm able to do 10-20x what I was able to do before AI, and I'm able to work at a breadth, depth, and level of polish that was previously impossible without mastering other diciplines (CSS, design, etc).
With AI, I've built several video games, my own SRS software, my own productivity software, and a business making $5M ARR and growing 70% month on month.
I've now done more hardware projects in the past eight months than I have in the last ten years. I might even start doing physics and bio (I did my undergrad in biochem) if I can find the time.
I'm also back to making movies. I was a major hobbyist filmmaker before the pandemic. I've shot dozens of films. They're extremely logistically complex and take so much time, and I was never capable of working in the high fantasy or sci fi genres I wanted to. Seedance is scratching so many of my itches it isn't even funny. It's like a gift from the gods.
I'm working more now than ever before, and I'm turning those hours of writing pointless glue code or testing new libraries and frameworks into actual productive work. Getting real work done.
2010-2020 felt like tech dystopia to me. Cellphone incrementalism, walled gardens, adtech, crypto, locked down devices. It was a pale shadow of the excitement of the early web. Personal websites, kids making Flash games, the indie web, p2p, people building stuff being 70% of the internet. For the first time in a long time, I feel extremely hopeful about the future and excited to work with tech again.
tl;dr - I can still write in cursive, but it turns out I just don't need to as I write my movie.
I do not get you folks who claim this is bad. This is magical.
Why do you (and lots of other people with your mindset) speak to other humans this way?
I post stuff I make on reddit and the teenagers (or grown men?) say things like, "your mother should have used a coat hanger if she knew you'd use AI". And it's almost a daily occurrence on social media.
I'm having a great time, and you don't have to join the club. Do whatever you want with your life. But stop talking to us like this.
I get that what I posted here is a critical refutation and not simply a neutral art or tech post about AI to a friendly audience. It invites critique and that's fine. But all the same, why do people go out of their way to frame it like this? To dunk on positive things and experiences people are having?
> But all the same, why do people go out of their way to frame it like this?
Because you (and your legion of likeminded folks on pseudosocial media) indeed sound like some washed-up, self-delusional salesman with a loaded gun to his head, desperately trying to land that sales pitch. And that's before we come to address all the usual cheap clichés, false equivalences, or if you wrote that slop yourself in the first place, etc.
I don't make comments like the appalling one you quote, and I'm sorry someone said that to you.
That said, I do hate genAI, and so have decent grounds for guessing why people would say things like that.
In my case, my employers have mandated LLM usage and all but banned actually programming, and I can see it rotting my ability to code and understand things.
That is extremely demoralizing, and I expect others going through it would be very likely to lash out in anger when they see people praising genAI as amazing and wonderful.
Distinct from that, those of us who have at one point supported ourselves in a purely artistic domain we love, such as music or illustration, are deeply saddened to see genAI destroying the opportunity to be paid to do those things we love. We hope that other humans will continue to invest their wealth into art made by humans, but clearly some are not, and that group may grow as the generators get better.
The pain of seeing the opportunities you had destroyed, both for yourself and for others, is made even more acute when you see someone shouting about how wonderful the tools causing the destruction are. Angry aggression is the expected result, we humans being what we are.
Finally, many of us in the anti-genAI camp are both able to see that the tools can be effective force-multipliers, but can also see the manifold negative externalities of genAI: on-demand propaganda, genAI psychosis, anonymous virtual rape (I don't know a better way to describe generating and distributing videos of actual humans doing sex acts they've never done), super-persuader scammers, potential massive job loss, concentration of social / political power into the GPU-holders' hands, accelerated climate change from the massive resource consumption, actual unaligned paperclip optimizers (see e.g. the HuggingFace hack, which is a textbook example of a monkeys-paw / paperclipper scenario), actually-effective mass surveillance, the increasing risk of bioterrorism at massive scale, and the looming possibility that an ASI evolved from these systems will actually bring us to "I Have No Mouth And I Must Scream" or extinction scenarios.
If you see all that, the urge to bully, insult, and denigrate people who are choosing all those consequences by using the things is very strong. Viewed from this angle, it's certainly hard not to conclude that those of us using genAI are being very, very selfish, choosing personal benefit over the collective good.
I feel guilty daily for not having the courage to refuse to use and accept the economic consequences (if I weren't a special-needs family's primary breadwinner, I would just refuse and accept the consequences).
Again, I'm sorry people are responding abusively to you, and I hope I will never react that way.
Have I helped you see why people might react cruelly to your comments?
I believe you. I'm just telling my own personal experience. What works for you doesn't seem to work for me. Maybe I'm doing something wrong? But also: I'm purely doing things for my own pleasure now, so whatever I enjoy more is what I'm free to do. Again though, that's just me.
I would absolutely believe people on the spectrum with hyperfocus-oriented ADHD (such as myself) might knock things out of the park with coding agents while everyone else is befuddled by their bizarre behavior. But then, I don't get the slow pace and fake drama of reality TV and professional wrestling, yet so many others do. Similarly, starting with British serial dramas like Survivors and Blake's Seven in the 80s, I really enjoy serialized dramas with an intricately constructed world. We are not all alike inside IMO, but that's 100% a good thing too.
This isn't UGC. I know a lot of studios from my filmmaking and 30% of my customers are spending thousands of dollars a month. It's got really good word of mouth and 89% M3 retention.
There are two films coming out in theaters this year that use ArtCraft.
It's also completely open source (the desktop app and the server).
There's another variant of ArtCraft that works with other subscriptions (Runway, Higgs, OpenArt, Comfy, etc.) that I'm testing with some artists: https://artcraftx.ai
Artists are fine tuning their own workflows on top of my own source software and then sharing their variants of ArtCraft.
I'm building OpenSourceRouter.com (soon) that will be an open source OpenRouter/FAL/RunPod/Featherless fine tunes. It already powers ArtCraft and is 90% complete to launch. Maybe in the future compute will be fully portable and you can just login to services with compute.
I'm also building a collaborative space for AI artists to share workflows, refs, fine tunes, network, and collaboratively build. This side of the business is already here and growing and I haven't built the infra for it yet. It's currently in Discord and an assortment of other places.
I'm going to make ArtCraft work with local models too. Comfy is not comfortable, and an all-rust/native, heavily opinionated product with lots of love from filmmakers and artists can do a better job than node graphs IMO.
I really care about these tools and I'm doing all of this from a place of passion. I'm not a mercenary trying to get rich quick. I'm here for the mission of empowering artists (and I count myself - this is a tool for me.)
It's growing really fast and I'm talking to bigger and bigger folks. It's kind of wild.
Hear hear. I do not understand the people who bemoan the loss of slide rules and books of trig tables in the era of graphing calculators.
I’m with you — I’m producing more and better art, with a deeper understanding of the art which comes from scale. My music is more innovative, my simulations are richer and more fun, I can wake up in the morning with a new idea and have a prototype in hours. It is magical.
Big difference is when you use a calculator or other tools, you still have to interpret and know what to enter.
With modern AI, you basically just ask for something and what used to be days worth of work magically appears, you didn't really think through a single thing when it comes to implementation, architecture etc unless you made a conscious effort to audit that.
We also generally still encourage students to learn how to do certain operations and what they mean without a calculator before letting them rely on it to make things quicker. If you never knew what a sin/cos was you'd never even know when to use those or why they even exist on a calculator.
AI is a step up where you literally don't have to know what those are, you could just take a picture and ask for the solution and it would do that all for you if you let it.
In reality modern AI is a massive test of willpower. You can completely let go of the reigns and it will happily do everything for you.
> Hear hear. I do not understand the people who bemoan the loss of slide rules and books of trig tables in the era of graphing calculators.
Why do you think so many people prefer to listen to live music over an album playing on a jukebox?
Live music doesn't have the polish and perfection of a studio recording, it's much less efficient to have to manually play instruments than to just hit "play" on a song. So why bother?
I think it's because we're looking at things wrong all over the place.
Similar to live music, I definitely prefer to read code written by a real human. I can get in their head, understand how they were reasoning to solve a problem, even if there were problems with their reasoning.
When I hear all the breathless takes about AI "productivity" improvements, my brain just circles aback to the old saw about activity over achievement.
What are your outputs? What is activity and what is achievement?
A business person brags that AI is able to generate a report for them that used to require lots of manual work linking spreadsheets. OK.
What was the value of the report? What was the goal? I like to follow this all the way down-
- Widget sales report
Why?
- To inform inventory and production decisions, as well as marketing
Why?
- So that the company can make money? So that the founders can be proud of their product? So that the product can improve lives in some way?
Why?
- So that the people working can have improved lives, and/or more status in society
Why?
- So that collectively by competing with each other in structured ways, humans as a whole thrive and benefit
Etc...
OK. Where does AI fit into this? What has it taken away from the human experience, and how had it affected all of these goals?
The goal isn't code. It isn't even the product that the code makes - it's the reward signaling that you get from cooperating and competing with other people.
When you short circuit your efforts at creation by using AI, it feels the same to me as saying you're putting on a live music show, but instead just hitting "play" on an album.
In my experience, students were actually pretty good about understanding this for themselves, at least in a security course I ran a survey in[0]. Unfortunately I didn’t get a statistically significant sample size (10/~60).
Another caveat is this was a graduate course, so I would expect more maturity and integrity on average than an undergrad course. But anecdotally I thought it was an interesting result.
That reminds me of "coding" in my Roaring Spring No.20165. Which is to say, my notebook. It was great for when I didn't have access to a computer. I remember the feeling of how cool it looked along side cartoons of monsters and cars crashing into the paper's edge.
After many decades of coding, probably 1M+ LOC, for real, I am purposely doing most of my work through coding agents to understand their capabilities. No stupid demand that I take a day off from AI is remotely helpful here.
But broadly my preliminary findings after a year or so:
1. AI gets you to that MVP 80% dopamine rush and then takes all that dopamine away if you lean into doing the last 20% with it. But if you try to do the last 20% yourself, you will experience the joy of agentic coding style and you will lose the ability to even produce dopamine. I use these things to build PoCs, not products.
2. The haters mostly have no idea how amazing getting to that 80% really is as a technological achievement and they are mostly stuck at Stochastic Parrot. And 3 days of the ungimped Fable demonstrated to me most of that remaining 20% is achievable, but not with gimping and/or production optimizations by the frontier labs.
3. If you want to see the promise and the limits of coding agents, you have to try to build something with them. But if you hate AI, you've already made up your mind and no amount of evidence will sway you, so don't bother.
4. These things are hideously power inefficient. And Moore's Law is out of steam. While we could stand up renewable and nuclear to power this, there are factions that hate one or the other more than they hate AI so good luck with that.
5. The San Francisco AGI cult is 100% unhelpful and the C-suite linkedin lunatics who take them seriously are even worse.
I'm reading a book written in the 90s but set in the 1980s where one of the assertions was that America is incompatible with technocracy. Prophetic methinks.
For my personal “project” I’ve been using chatgpt to teach me from math (linear algebra) to quantum computing - asked it to create a big list of prompts optimal for single threads, for me to learn with Socratic method.
I plan to write my notes as a blog for each thread in my own words.
I think this is a good use of AI for personal goals, it’s the best learning experience for me so far. I’ve never been able to focus on lecture videos
Reminds me of when we were admonished to use a "real" compiled language rather than an interpreted, GC'd one, lest we lose our ability to reason about the system, cache, and memory.
Which reminds me of when we were admonished to continue using assembly, lest our "real" skills atrophy from too much cushy C development.
Which reminds me of when we were admonished to keep our scope and probe skills sharp, and read and understand the schematics of whatever computer we were on, lest we forget how to diagnose faults in its components.
Which reminds me of when we were admonished to practice calculating in our heads and keeping slide rule skills sharp, lest they atrophy to a calculator.
You don't need to "keep practicing" in order to maintain your understanding.
The best engineers know how our systems work down to more levels, but for average developers producing average software with average impact, average output is acceptable.
>Reminds me of when we were admonished to use a "real" compiled language rather than an interpreted, GC'd one, lest we lose our ability to reason about the system, cache, and memory.
And they were right. People used in interpreted and GC ones lost most if not all the ability to reason about the system, cache, and memory.
Today's bloated "1GB RAM for a small form app" apps are the results of this.
>You don't need to "keep practicing" in order to maintain your understanding.
> People used in interpreted and GC ones lost most if not all the ability to reason about the system, cache, and memory.
It's not that they lost it, but rather that they never learned it. I didn't lose my ability to reason about those things during the 4 years where I worked in Python and Java.
And once again, for average software devs that's fine. Most non-critical software that doesn't perform very complex tasks can be optimized for time-to-market, which of course loses in other areas the business doesn't care as much about (such as bloat).
I get the impression that analogies like this are trying to disguise a moral judgement as a practical one.
It's clearly absurd to abandon GC languages or to require everyone who uses one to learn assembly. Yes, it could be beneficial to learn the lowest levels of the machine, but with an opportunity cost. Advocates of many such principles are often those who have already paid that opportunity cost and consequently see no downside.
There has always been a known trade off with how much you need to know about a system against the practical ability to make something that someone uses.
If we talk directly to the practicality issue, what happens when AI can rewrite a program written in a highly expressive, but inefficient, language? A future AI might be able to turn a bloated prototype into a sleek, high performance, program where every byte is justified.
Then it would be absolutely true that a programmer who did not know the low level intricacies would not know how to do that for themselves, but would they need to? If they can make the things that they want to make and those things do not have obvious limitations, they could be learning a skill more relevant to the world that they live in.
By all means, learn assembly if you think it is interesting. I have learned Asm (and indeed written assemblers) for a variety of systems. I won't hold it against someone if they spend the time figuring out cross entropy instead.
This is a paradigm shift. If you know basic arithmetic, a calculator doesn't stop you from knowing how to add or multiply. But AI definitely stops you from learning how to reason. So it's not the same at all.
Average output is acceptable at work, sure, and if all you want is a paycheck then yeah I agree, make Claude write all your code, who gives a damn. This advice is for people who want to learn
Reminds of the time when we had to plant our own apple trees instead of buying them from the market, and to get up early at 1 AM to milk the oranges instead of just buying juice
Oh man, I help host a monthly AI meetup called AI Friday: https://aifri.day
My heart sank when I saw this post. I thought we’d created enemies!
No conflict here, though.
Personal example... A few weeks ago, I realized that with all these agents running and all the things I could be working on, I was spending my days managing shallow pings instead of going deep.
I have AI agents managing SEO and ad spend. I used to spend a little time every day reading what each had to say. Now each gives me one detailed report per week, with links to the underlying reports and specific things I need to look at. I block 90 minutes for each one, on different days, and use that time to go deep. It’s helped a lot.
I would actually really recommend people who are using coding agents full time to make AI brakes for your weekend projects. Think of it like a small break to take the time to reconnect to the code and keep your brain muscles intact, a bit like a professional cyclist going for a casual ride in the forest to experience what they don’t experience anymore in their professional life.
I see where you're coming from, but I can't help being reminded of my youth when teachers would admonish me "YoU wOn'T aLwaAYs haVe A CalCulator in yOuR poCKet!"
Now the entire world has a calculator in their pocket, and all that wrote memorization not only made me hate math as a child, it was a giant waste of time.
I suspect that one day we'll look back on advice like no-AI Fridays exactly as modern readers would react to "Punch-card only Tuesdays" in a modern enterprise.
How much rote memorization did it take to harm you? the addition of 1 digit numbers is kind of mandatory to function in everyday life, the multiplication table is at most (8*9)/2 36 facts.
> Now the entire world has a calculator in their pocket, and all that wrote memorization not only made me hate math as a child, it was a giant waste of time.
I dunno, I think this example kind of works in the opposite direction. If we're talking about memorising tables of logarithms or something, then yeah that was a complete waste of time. But a lot of the stuff kids would like to skip because calculators can do it is the absolute basics. IMO it is still important to have a solid grasp of arithmetic, useful to have enough trigonometry that you intuitively understand where it can be useful, and so on. (Much more so if you want to leave open the possibility of studying further mathematics.) And for people to whom these things don't come naturally, you really need to drill it a bit in order to develop the conceptual understanding.
I don't know exactly where this does and doesn't apply to programming/CS, though. I would like to think that a lot of the fundamentals will remain useful, but maybe that will only be true for a few specialised workers and self-motivated hobbyists.
As someone who has made a career by having unique insights into solutions for software problems, I can confidently says it's not the same as the calculator. AI is bad at gathering context from an in-person customer meeting, linking their experience to fundamental structural problem with the data model and formulating a solution. You know how you know to make that insight? You have to have read and internalized the code. The easiest way to do that is by having written it yourself by hand or direct instruction to the AI. This exact scenario has played out for me more times than I can count, and it's why I find so little value in using AI.
More like: take some time without your phone. Something I assume everybody here would agree is a sane thing to do.
Not using a calculator as a kid for math is pretty important, because the goal is to develop intuition. Same for software, playing music, learning a language. You have to keep some level of friction to develop and keep your intuition
hah. try go hiking for couple days in mountains. no cellular, no charging, flaky gps, vhf/uhf radio don't work and your only hope is that you are prepared enough to survive getting out of there .
I don't even about sailing 100+nm off shore.
people assume every point of earth is somewhy equivalent to their porch.
Not entirely sure what point you're trying to make, so maybe this is not relevant to it, but, in 2026, satellite emergency support makes it seem irresponsible to not take your iPhone if you value your life or have other people in your life who would miss you if you were gone.
For me its quite the opposite. My employer is too afraid to commit to any of it. Meanwhile, I've built more stable software for myself and my personal computing needs of higher quality and in faster turnaround times.
I've also seen how insanely good at doing a total rewrite / modernization the models can be.
I am currently on an obsessive coding spree on my side projects where I only use AI to ask questions like I used to do on Stack Overflow. At work it’s nearly 100% AI because quite honestly the AI is more thorough and faster than humanly possible. I don’t get my coding itch scratched at work anymore, so my off work coding has become a necessity for me. At first it felt pointless, I know I could just write a prompt and get the code written in a minute. But now, as the blog post says, I feel my ability degrading and doing things by hand has become really important for me to try to stop that.
The question however is whether the new generation of devs will need to be able to write and review code by hand or AI will just do it so much better than humans that it will make no sense to do that, and what the humans will need to focus on instead. I fear this thinking that humans must understand the code and keep the AI from making a mess may be a relic of our pre AI brains, and soon if not now already it’s quite the opposite.
It takes an agent on the order of 100 billion operations to produce a single token. A person (generously) taking just a second for a single mental addition or multiplication would take tens of thousands of years to write just a single line of code.
And what am I supposed to do when I hit a wall solving a problem? Just sleep on it?
Sarcasm aside, I think there's a real phenomenon of AI removing the experience of getting stuck on a problem (where often I'll wake up the next day with a perfect solution in my head ready to implement). The same way social media removed the experience of "being bored".
Giving yourself time to be idle and let a problem marinate is often where the breakthroughs and good ideas come from. If taste and human judgment are what separate useful AI output from slop, we need to give our brains time to cook instead of constantly bouncing between agents.
The technique I'm following is: If it's easier to explain it to an LLM, have the LLM do it. If it's easier to do it myself, I do it.
This is useful when iterating on UI, where I need to make a few minor changes; or when I need to "explore" to understand how a query or data model needs to work.
Otherwise, I don't need to dedicate a day to doing things by hand; I just do a few things by hand every day.
I just cleaned my cats litter box and that was cooler than AI.
I have no idea how people can spend 40 hours a week interacting with those sycophantic random number generators, its like having the most annoying linkedin influencer you can imagine DM'ing you all day.
My old workplace did the opposite: we had All AI Fridays where you had carte blanche to build out AI tooling or just work on AI stuff as long as it's vaguely related to your job.
Maybe this is a useful way to address maladaptive llm use for some, but it seems to focus the discussion on either-or (using llms) as opposed to how (to use them).
For things you want to learn: Are you asking to have something solved, or are you asking for information so you can solve it?
There may be detrimental learning effects even with asking llms for information, but I think there's a fundamental difference in addressing it as an intern/slave vs teacher/tutor.
i inadvertently do this at work anyway. i burn through my monthly token budget on boring-but-valuable tasks, and then i spend the rest of the month doing stuff by hand.
When I was 9 or 10 I flew somewhere with my family for the first time. On landing, the pilot bounced the plane off the runway. While taxiing he came on the intercom and apologized for the "slightly rough" landing, explaining that they usually use <fragmented memory> some autopilot to land/smooth the landing and that to keep qualified they have to do it manually every so often.
So anyways, I fully embrace and support the idea of bouncing your code off the compiler every Friday. It's what I do and I can really tell those skills need practice even after just a week.
I don't know about Fridays but I like spending some significant amount of time full-vibing then some significant amount of time full-manual, coming to terms with what has been done-but-not-pushed, figuring out how I want the code to actually work and look, with roughly zero regard to how much transformation or starting-anew it will take from what has been vibed - easy come, easy go. Then once I really know what exactly I want, either transform the vibed code into it, or delete and start over (maybe manually, maybe prompting).
Nobody should be in a position where they need to clear something like this with their boss. Now, I'm certain people are but that just means those of us with seniority and in leadership have failed those people.
We need No AI Hours everyday. We need this kind of time every day to clear our minds, reflect on what AI has brought us, and consider whether it's necessary to address the issue of technical debt.
100%. Making it a whole day makes it easier to sell and stick to, that's it. It's just brutal to have a whole day when you need to focus and not just mindlessly prompt.
Sounds good. They're encouraging others to join in but it'd be great to hear if this actually has an impact so far and how its worked out for HTMX. I imagine there are some interesting anecdotes about what actually happens when teams do this.
[Edit] just seen that OP is pitching this rather than practicing it. I wonder if any team actually do have a practice like this...
The cognitive debt accusation is dumb. Implying there is some sort of cognitive deterioration when it is merely a label used by the authors to describe lower cognitive engagement while using AI and some persistence of that lower engagement when returning to the task without it … which is predictably going to be the case when you’re offloading work.
If your plan is to offload a task to AI, do you really need to dedicate your attention to it?
I find it interesting when people say, “I need time away from the AI to think.” I use AI for development but I would not say I’ve offloaded my thinking to AI. AI is great for pushing text but I haven’t found its system design skills to be that good.
To me the design has always felt like the thinking though.
I felt the same until I actually tried to stop for a week or so. And I could feel the constant temptation and anxiety of not having the coding agents as a sort of safety net. Taking a clean brake made me realize how much I was losing my skills and connection to the actual implementation, not just the high level pieces of it. I also didn’t realize at the time how much of my design thinking is done while working on code, and all the constant small decisions that are actually impactful
Great initiative. I was expecting something like this to happen. I think a good addition to the website can be success stories people share from this approach. Sending a link this to a manager is not very convincing, but managers respond very well to social proof.
This is as silly as saying you shouldn’t drive your car on Friday’s and walk to work instead.
Why only Friday? What does “doing things the old way” one day a week prove or benefit when you spend 90% of your time working with AI?
Tech people who have their entire identities wrapped up in being smart or the best engineer in the room have suddenly learned that intelligence is now a commodity and a cheap one at that.
Objectively, it actually would be beneficial for most people to walk to work once a week…
Also, being intelligent is still important. Enabling the dumbest people in the room to dump their tidal wave of slop on others isn’t really a positive.
"Not too much AI" or "Just enough AI" also, if you feel the need for a No AI Friday it could mean you also need to just use less AI, or try to only use it for the correct things so you don't lose visibility, for example don't let it make decisions for you, have it present the decisions.
Choosing from a set of decisions the agent presents you might feel like you are making a decision, but you’ll never make the one that wasn’t presented.
I’m not going to go that far (you do you), but reading the source code seems like a great way to remain in touch with the implementation. On my hobby project, I do a lot of reading code, asking the AI pointed questions about how things work, and asking it to clean up messes.
So you’re left with the least interesting tasks? Not trying to be too snarky, or too much of a luddite, but that would be a killer for me, motivationally.
Maybe we just disagree about what's fun? Some people like to build things by hand and other people enjoy working with power tools.
I'm having a great time building personal web apps with exactly the features I want, and I also enjoy refactoring the source code to get closer to how I would ideally write it. I can use a coding agent to do both. What's not to like?
I stopped editing source code by hand as a self-imposed limitation to see how far I could go by prodding AI to do things my way, and it turns out it's pretty far, so I'm sticking with it.
Which isn't to say that writing code by hand can't be fun, too, under the right circumstances.
It also helps that I’m retired and this is hobby programming. If you’re being forced into it by circumstances at work, that wouldn’t be nearly as fun.
This will be the pattern only the "privileged" workers will be able to afford.
Not using AI is similar to giving employees 20% of their time for any non-sprint work.
This privilege of No AI usage a day will lower the productivity, and rarely any companies will allow it (if they track your usage, which I am sure they do)
> Study after study shows that using LLMs can cause you to accumulate cognitive debt, make you less engaged with your work, negatively impact your critical thinking abilities, and hamper your skill formation.
False
I've learned more, I am more engaged, it has positively impacted my critical thinking abilities and has increased my skill formation
> ...cause you to accumulate cognitive debt, make you less engaged with your work, negatively impact your critical thinking abilities, and hamper your skill formation.
My employer gifted me all that years before LLMs were released.
Anecdotal example but I know plenty of students (Including my own brother) who graduated recently who did it almost entirely with AI (unfortunately at the request of their own professors) who can't answer the most basic of questions that I would have expected of sometimes even first year CS students in the days before AI was a thing, which is why I encourage them to build projects without it.
Personally, I'm starting to think that in university and education systems, all projects should now be done in class and in computer labs/monitored study halls because we can't trust anyone to not use AI anymore.
For my project I use Gemini/Google's AI mode to great effect. Quick answers to be reminded about syntax, and to get help when stuck. It actually enhances the flow state. Plus it's all free.
We used to call it "seeing through the code". The moment when you're no longer looking at text on a page, you're seeing what it does as you scroll through it.
What you need most, to get to that place, is to simply spend a lot of time writing and reading code until it becomes second nature. Something like 10,000 hours.
Once you've done that, reading an LLM's code is trivial. But letting it write most of the code, without reviewing it line by line, causes rapid brain atrophy. Because to really be in the zone, you need to keep everything relevant in your head. Which means it has to be in there in the first place.
As someone who gets "in the zone" like the person you're replying to, I just don't get how this could possibly be the same thing. "Coming back later" doesn't make any sense when you're in that state. Stopping and waiting is a killer. You're just constantly doing things without necessarily consciously thinking about it.
Do you mean you're switching continually between the twelve sessions? Or literally "coming back" in a different sort of operating mode?
The closest thing to it for me using AI tools is specifically not running multiple "agents" but making targeted changes over and over and over but with a 10x+ increase in typing-vs-change ratio. (This also rules out things like Opus let alone Fable as they're just too slow and do too much at once.)
I point at the violins. I gesture the drums upwards. Sometimes I make a big broad gesture that encompasses the entire orchestra.
My agents are tasked, in part, with explaining the context with mockups and examples I can poke at for each job.
I have found in the last year I "get more done" but I learn less. Whats even worse is if I burn though my token quota I am less likely(and less able) to "switch to manual". LLMs make an ocean of decisions and assumptions very quickly and if you don't slow it down very intentionally (and lose the promise of the AI providers are pitching) you will quickly not have a good idea of how systems work.
My actual coding ability is probably lower, but I’ve learned so much about music theory and production, CAD, and statistics. All because my hobby projects can go so much further than before.
How I experience it at least, there is no free lunch. Something that gives you knowledge and experience will almost by definition be painful and uncomfortable.
We use AI because it is faster, but that speed has a consequence.
I don’t expect it to be any different with AI, but YMMV.
OTOH, the output might stick?
Then the question becomes do you value knowledge for its own sake or for what it enables you to do? I don't have an answer.
I worked on six nines active active payments infra that handled billions in daily transaction volume. I've built laser projection systems to play interactive games on skyscrapers. That ability is not atrophying.
I'm able to do 10-20x what I was able to do before AI, and I'm able to work at a breadth, depth, and level of polish that was previously impossible without mastering other diciplines (CSS, design, etc).
With AI, I've built several video games, my own SRS software, my own productivity software, and a business making $5M ARR and growing 70% month on month.
I've now done more hardware projects in the past eight months than I have in the last ten years. I might even start doing physics and bio (I did my undergrad in biochem) if I can find the time.
I'm also back to making movies. I was a major hobbyist filmmaker before the pandemic. I've shot dozens of films. They're extremely logistically complex and take so much time, and I was never capable of working in the high fantasy or sci fi genres I wanted to. Seedance is scratching so many of my itches it isn't even funny. It's like a gift from the gods.
I'm working more now than ever before, and I'm turning those hours of writing pointless glue code or testing new libraries and frameworks into actual productive work. Getting real work done.
2010-2020 felt like tech dystopia to me. Cellphone incrementalism, walled gardens, adtech, crypto, locked down devices. It was a pale shadow of the excitement of the early web. Personal websites, kids making Flash games, the indie web, p2p, people building stuff being 70% of the internet. For the first time in a long time, I feel extremely hopeful about the future and excited to work with tech again.
tl;dr - I can still write in cursive, but it turns out I just don't need to as I write my movie.
I do not get you folks who claim this is bad. This is magical.
I post stuff I make on reddit and the teenagers (or grown men?) say things like, "your mother should have used a coat hanger if she knew you'd use AI". And it's almost a daily occurrence on social media.
I'm having a great time, and you don't have to join the club. Do whatever you want with your life. But stop talking to us like this.
I get that what I posted here is a critical refutation and not simply a neutral art or tech post about AI to a friendly audience. It invites critique and that's fine. But all the same, why do people go out of their way to frame it like this? To dunk on positive things and experiences people are having?
Because you (and your legion of likeminded folks on pseudosocial media) indeed sound like some washed-up, self-delusional salesman with a loaded gun to his head, desperately trying to land that sales pitch. And that's before we come to address all the usual cheap clichés, false equivalences, or if you wrote that slop yourself in the first place, etc.
That said, I do hate genAI, and so have decent grounds for guessing why people would say things like that.
In my case, my employers have mandated LLM usage and all but banned actually programming, and I can see it rotting my ability to code and understand things.
That is extremely demoralizing, and I expect others going through it would be very likely to lash out in anger when they see people praising genAI as amazing and wonderful.
Distinct from that, those of us who have at one point supported ourselves in a purely artistic domain we love, such as music or illustration, are deeply saddened to see genAI destroying the opportunity to be paid to do those things we love. We hope that other humans will continue to invest their wealth into art made by humans, but clearly some are not, and that group may grow as the generators get better.
The pain of seeing the opportunities you had destroyed, both for yourself and for others, is made even more acute when you see someone shouting about how wonderful the tools causing the destruction are. Angry aggression is the expected result, we humans being what we are.
Finally, many of us in the anti-genAI camp are both able to see that the tools can be effective force-multipliers, but can also see the manifold negative externalities of genAI: on-demand propaganda, genAI psychosis, anonymous virtual rape (I don't know a better way to describe generating and distributing videos of actual humans doing sex acts they've never done), super-persuader scammers, potential massive job loss, concentration of social / political power into the GPU-holders' hands, accelerated climate change from the massive resource consumption, actual unaligned paperclip optimizers (see e.g. the HuggingFace hack, which is a textbook example of a monkeys-paw / paperclipper scenario), actually-effective mass surveillance, the increasing risk of bioterrorism at massive scale, and the looming possibility that an ASI evolved from these systems will actually bring us to "I Have No Mouth And I Must Scream" or extinction scenarios.
If you see all that, the urge to bully, insult, and denigrate people who are choosing all those consequences by using the things is very strong. Viewed from this angle, it's certainly hard not to conclude that those of us using genAI are being very, very selfish, choosing personal benefit over the collective good.
I feel guilty daily for not having the courage to refuse to use and accept the economic consequences (if I weren't a special-needs family's primary breadwinner, I would just refuse and accept the consequences).
Again, I'm sorry people are responding abusively to you, and I hope I will never react that way.
Have I helped you see why people might react cruelly to your comments?
https://news.ycombinator.com/item?id=49500019
I mean, you do you I guess..
https://youtu.be/kzvQMdg66Go (Now with full animation control, not just blocking and posing.)
This isn't UGC. I know a lot of studios from my filmmaking and 30% of my customers are spending thousands of dollars a month. It's got really good word of mouth and 89% M3 retention.
There are two films coming out in theaters this year that use ArtCraft.
It's also completely open source (the desktop app and the server).
There's another variant of ArtCraft that works with other subscriptions (Runway, Higgs, OpenArt, Comfy, etc.) that I'm testing with some artists: https://artcraftx.ai
Artists are fine tuning their own workflows on top of my own source software and then sharing their variants of ArtCraft.
I'm building OpenSourceRouter.com (soon) that will be an open source OpenRouter/FAL/RunPod/Featherless fine tunes. It already powers ArtCraft and is 90% complete to launch. Maybe in the future compute will be fully portable and you can just login to services with compute.
I'm also building a collaborative space for AI artists to share workflows, refs, fine tunes, network, and collaboratively build. This side of the business is already here and growing and I haven't built the infra for it yet. It's currently in Discord and an assortment of other places.
I'm going to make ArtCraft work with local models too. Comfy is not comfortable, and an all-rust/native, heavily opinionated product with lots of love from filmmakers and artists can do a better job than node graphs IMO.
I really care about these tools and I'm doing all of this from a place of passion. I'm not a mercenary trying to get rich quick. I'm here for the mission of empowering artists (and I count myself - this is a tool for me.)
It's growing really fast and I'm talking to bigger and bigger folks. It's kind of wild.
I’m with you — I’m producing more and better art, with a deeper understanding of the art which comes from scale. My music is more innovative, my simulations are richer and more fun, I can wake up in the morning with a new idea and have a prototype in hours. It is magical.
With modern AI, you basically just ask for something and what used to be days worth of work magically appears, you didn't really think through a single thing when it comes to implementation, architecture etc unless you made a conscious effort to audit that.
We also generally still encourage students to learn how to do certain operations and what they mean without a calculator before letting them rely on it to make things quicker. If you never knew what a sin/cos was you'd never even know when to use those or why they even exist on a calculator.
AI is a step up where you literally don't have to know what those are, you could just take a picture and ask for the solution and it would do that all for you if you let it.
In reality modern AI is a massive test of willpower. You can completely let go of the reigns and it will happily do everything for you.
Why do you think so many people prefer to listen to live music over an album playing on a jukebox?
Live music doesn't have the polish and perfection of a studio recording, it's much less efficient to have to manually play instruments than to just hit "play" on a song. So why bother?
I think it's because we're looking at things wrong all over the place.
Similar to live music, I definitely prefer to read code written by a real human. I can get in their head, understand how they were reasoning to solve a problem, even if there were problems with their reasoning.
When I hear all the breathless takes about AI "productivity" improvements, my brain just circles aback to the old saw about activity over achievement.
What are your outputs? What is activity and what is achievement?
A business person brags that AI is able to generate a report for them that used to require lots of manual work linking spreadsheets. OK.
What was the value of the report? What was the goal? I like to follow this all the way down-
- Widget sales report
Why?
- To inform inventory and production decisions, as well as marketing
Why?
- So that the company can make money? So that the founders can be proud of their product? So that the product can improve lives in some way?
Why?
- So that the people working can have improved lives, and/or more status in society
Why?
- So that collectively by competing with each other in structured ways, humans as a whole thrive and benefit
Etc...
OK. Where does AI fit into this? What has it taken away from the human experience, and how had it affected all of these goals?
The goal isn't code. It isn't even the product that the code makes - it's the reward signaling that you get from cooperating and competing with other people.
When you short circuit your efforts at creation by using AI, it feels the same to me as saying you're putting on a live music show, but instead just hitting "play" on an album.
Another caveat is this was a graduate course, so I would expect more maturity and integrity on average than an undergrad course. But anecdotally I thought it was an interesting result.
[0] https://exploiting.systems/posts/2026-05-17-learning-focused...
But broadly my preliminary findings after a year or so:
1. AI gets you to that MVP 80% dopamine rush and then takes all that dopamine away if you lean into doing the last 20% with it. But if you try to do the last 20% yourself, you will experience the joy of agentic coding style and you will lose the ability to even produce dopamine. I use these things to build PoCs, not products.
2. The haters mostly have no idea how amazing getting to that 80% really is as a technological achievement and they are mostly stuck at Stochastic Parrot. And 3 days of the ungimped Fable demonstrated to me most of that remaining 20% is achievable, but not with gimping and/or production optimizations by the frontier labs.
3. If you want to see the promise and the limits of coding agents, you have to try to build something with them. But if you hate AI, you've already made up your mind and no amount of evidence will sway you, so don't bother.
4. These things are hideously power inefficient. And Moore's Law is out of steam. While we could stand up renewable and nuclear to power this, there are factions that hate one or the other more than they hate AI so good luck with that.
5. The San Francisco AGI cult is 100% unhelpful and the C-suite linkedin lunatics who take them seriously are even worse.
I'm reading a book written in the 90s but set in the 1980s where one of the assertions was that America is incompatible with technocracy. Prophetic methinks.
Which reminds me of when we were admonished to continue using assembly, lest our "real" skills atrophy from too much cushy C development.
Which reminds me of when we were admonished to keep our scope and probe skills sharp, and read and understand the schematics of whatever computer we were on, lest we forget how to diagnose faults in its components.
Which reminds me of when we were admonished to practice calculating in our heads and keeping slide rule skills sharp, lest they atrophy to a calculator.
You don't need to "keep practicing" in order to maintain your understanding.
The best engineers know how our systems work down to more levels, but for average developers producing average software with average impact, average output is acceptable.
And they were right. People used in interpreted and GC ones lost most if not all the ability to reason about the system, cache, and memory.
Today's bloated "1GB RAM for a small form app" apps are the results of this.
>You don't need to "keep practicing" in order to maintain your understanding.
You 100% need to.
It's not that they lost it, but rather that they never learned it. I didn't lose my ability to reason about those things during the 4 years where I worked in Python and Java.
And once again, for average software devs that's fine. Most non-critical software that doesn't perform very complex tasks can be optimized for time-to-market, which of course loses in other areas the business doesn't care as much about (such as bloat).
When that's not the case, you hire a specialist.
It's clearly absurd to abandon GC languages or to require everyone who uses one to learn assembly. Yes, it could be beneficial to learn the lowest levels of the machine, but with an opportunity cost. Advocates of many such principles are often those who have already paid that opportunity cost and consequently see no downside.
There has always been a known trade off with how much you need to know about a system against the practical ability to make something that someone uses.
If we talk directly to the practicality issue, what happens when AI can rewrite a program written in a highly expressive, but inefficient, language? A future AI might be able to turn a bloated prototype into a sleek, high performance, program where every byte is justified.
Then it would be absolutely true that a programmer who did not know the low level intricacies would not know how to do that for themselves, but would they need to? If they can make the things that they want to make and those things do not have obvious limitations, they could be learning a skill more relevant to the world that they live in.
By all means, learn assembly if you think it is interesting. I have learned Asm (and indeed written assemblers) for a variety of systems. I won't hold it against someone if they spend the time figuring out cross entropy instead.
In the snow.
Uphill.
Both ways.
My heart sank when I saw this post. I thought we’d created enemies!
No conflict here, though.
Personal example... A few weeks ago, I realized that with all these agents running and all the things I could be working on, I was spending my days managing shallow pings instead of going deep.
I have AI agents managing SEO and ad spend. I used to spend a little time every day reading what each had to say. Now each gives me one detailed report per week, with links to the underlying reports and specific things I need to look at. I block 90 minutes for each one, on different days, and use that time to go deep. It’s helped a lot.
Now the entire world has a calculator in their pocket, and all that wrote memorization not only made me hate math as a child, it was a giant waste of time.
I suspect that one day we'll look back on advice like no-AI Fridays exactly as modern readers would react to "Punch-card only Tuesdays" in a modern enterprise.
I dunno, I think this example kind of works in the opposite direction. If we're talking about memorising tables of logarithms or something, then yeah that was a complete waste of time. But a lot of the stuff kids would like to skip because calculators can do it is the absolute basics. IMO it is still important to have a solid grasp of arithmetic, useful to have enough trigonometry that you intuitively understand where it can be useful, and so on. (Much more so if you want to leave open the possibility of studying further mathematics.) And for people to whom these things don't come naturally, you really need to drill it a bit in order to develop the conceptual understanding.
I don't know exactly where this does and doesn't apply to programming/CS, though. I would like to think that a lot of the fundamentals will remain useful, but maybe that will only be true for a few specialised workers and self-motivated hobbyists.
Not using a calculator as a kid for math is pretty important, because the goal is to develop intuition. Same for software, playing music, learning a language. You have to keep some level of friction to develop and keep your intuition
I don't even about sailing 100+nm off shore.
people assume every point of earth is somewhy equivalent to their porch.
I've also seen how insanely good at doing a total rewrite / modernization the models can be.
How is this supposed to work? Like emulate an agent with my brain and write down the output?
It takes an agent on the order of 100 billion operations to produce a single token. A person (generously) taking just a second for a single mental addition or multiplication would take tens of thousands of years to write just a single line of code.
Sarcasm aside, I think there's a real phenomenon of AI removing the experience of getting stuck on a problem (where often I'll wake up the next day with a perfect solution in my head ready to implement). The same way social media removed the experience of "being bored".
I also support No GenAI every day ending in a Y, but I support this, too.
Giving yourself time to be idle and let a problem marinate is often where the breakthroughs and good ideas come from. If taste and human judgment are what separate useful AI output from slop, we need to give our brains time to cook instead of constantly bouncing between agents.
I’d make a joke about there not being a word for the reverse of plagiarism, except there is one: lying.
This is useful when iterating on UI, where I need to make a few minor changes; or when I need to "explore" to understand how a query or data model needs to work.
Otherwise, I don't need to dedicate a day to doing things by hand; I just do a few things by hand every day.
I have no idea how people can spend 40 hours a week interacting with those sycophantic random number generators, its like having the most annoying linkedin influencer you can imagine DM'ing you all day.
TODO: insert something about open source, beer, free beer, someone else's beer, trading beer, subgenius, slack etc etc
For things you want to learn: Are you asking to have something solved, or are you asking for information so you can solve it?
There may be detrimental learning effects even with asking llms for information, but I think there's a fundamental difference in addressing it as an intern/slave vs teacher/tutor.
So anyways, I fully embrace and support the idea of bouncing your code off the compiler every Friday. It's what I do and I can really tell those skills need practice even after just a week.
[Edit] just seen that OP is pitching this rather than practicing it. I wonder if any team actually do have a practice like this...
That is not minimising.
If your plan is to offload a task to AI, do you really need to dedicate your attention to it?
To me the design has always felt like the thinking though.
Could be another case for four-day work weeks, especially if AI is supposedly improving productivity during those four days a week.
And its four preprints and one published one with an extremely weak experiment.
But seriously who offloads decision-making to LLMs? Thats the main thing they can't do proper.
Why only Friday? What does “doing things the old way” one day a week prove or benefit when you spend 90% of your time working with AI?
Tech people who have their entire identities wrapped up in being smart or the best engineer in the room have suddenly learned that intelligence is now a commodity and a cheap one at that.
Also, being intelligent is still important. Enabling the dumbest people in the room to dump their tidal wave of slop on others isn’t really a positive.
I'm having a great time building personal web apps with exactly the features I want, and I also enjoy refactoring the source code to get closer to how I would ideally write it. I can use a coding agent to do both. What's not to like?
I stopped editing source code by hand as a self-imposed limitation to see how far I could go by prodding AI to do things my way, and it turns out it's pretty far, so I'm sticking with it.
Which isn't to say that writing code by hand can't be fun, too, under the right circumstances.
It also helps that I’m retired and this is hobby programming. If you’re being forced into it by circumstances at work, that wouldn’t be nearly as fun.
Reading code is way less enjoyable than writing it
Not using AI is similar to giving employees 20% of their time for any non-sprint work.
This privilege of No AI usage a day will lower the productivity, and rarely any companies will allow it (if they track your usage, which I am sure they do)
The end goal of every AI initiative is to lower headcount, if your employer is forcing AI usage you are digging your own grave.
> Study after study shows that using LLMs can cause you to accumulate cognitive debt, make you less engaged with your work, negatively impact your critical thinking abilities, and hamper your skill formation.
False
I've learned more, I am more engaged, it has positively impacted my critical thinking abilities and has increased my skill formation
For me it's AI 24/7 even on sundays
If it’s just your own datapoint, is it difficult to self assess?
My employer gifted me all that years before LLMs were released.