It is opposite for me than what the author describes. With AI I can explore new areas far more effectively. For example, I have never been great with front end web stuff, but asking LLMs to generate something and then looking at the code they built is pretty cool.
Of course there are errors and gaffes and weird bits all over, but that helps keep me in the loop and understanding it better.
I can now (more) boldly go where I feared to tread before.
titularcomment
With all due respect, what you're describing doesn't quite spell learning in my opinion. Sure, its better to read the code of an application you envisioned to learn a programming language rather than an unrelated convulated program; but you're still offloading your thought process to the LLM. You're not quite seeing why it chose A instead of B, and why it uses that struct instead of the other. And honestly, the LLM doesn't quite know why it chose that either.
Scubabear68
I am describing a problem and seeing how the LLM solves it (or how it goes off the rails). I am often exposed to techniques new to me. I call that learning.
To be clear, I am closing in on 60 years old. I’ve written a lot
of code in a lot of languages. I have not bothered to check every fad that has come down in front end web and Javascript for the past umpteen years. Front end hasn’t been my thing. With current AI, I can leap frog a lot of bull crap I don’t care about and see what at least the AI thinks is good. I still retain a brain, I can still investigate what it did and form an opinion about it.
I treat the AI stuff as idiot
savants that can make really nice stuff or go totally off in the weeds. But it is more than good enough to help me be a better me at what I do.
2muchcoffeeman
> “ honestly, the LLM doesn't quite know why it chose that either.”
Uh, have you asked it?
These sorts of articles are funny. Like the author assumed that you have to use an LLM in a certain way.
“I can just ask the LLM and there it is! I didn’t learn anything! :(“
You can interrogate the LLM. You can ask it all sorts of things. Tell it to justify its decisions. Tell it to give you supporting evidence. Tell it to give you options. Tell it to explain things in detail.
I’ve been using it as a learning tool and it’s quite effective.
Granted the stuff I don’t care about really does get worse. If I have a config typo I just go “There’s a typo in blah. Find it and fix it.” Or setting up the boiler plate for a project. I’m getting quite useless at that.
But it doesn’t have to be any particular way. It’s a tool. Exploit it.
beej71
I use it as a very effective learning tool, as well. To test my learning, I make sure I can code it myself after that.
phoghed
Nowhere did they say it’s learning. They said they can do more stuff now, it’s exploration.
titularcomment
The original article talks about 'savvy', which is to me an essence of learning. If 'exploration' is looking at walls of code you don't understand but the linter makes them look cool, you don't need LLMs for that.
nobodywillobsrv
I see this response a lot. What part of the thinking are you thinking is actually offloaded?
Writing things down and then arguing with an agent that constantly gets things wrong is just a new kind of Rubber Duck.
Are Rubber Ducks offloading thinking?
These are more like search plus memory tools.
Memory loss is real but that has been happening for me way before agents.
titularcomment
This came out a bit longer than I expected.
Depends on your usage. A rubber duck won't upheave the entire history of written human knowledge to reach for your answer. It doesn't have inherent bias, and its purpose could also be handled by a wall, or the air infront of you if you can stretch your mind a little. What I'm going for here is that as you said, its just a hack to make your brain look for answers in a more efficient manner, by clearly underlining your thought process out loud, tricking yourself into conversation mode, yada yada. Its still you doing all the work, you're just being smarter about it.
A LLM is more akin to a senior pair programmer. It may seem like a rubber duck on the surface, in the sense that you're having a conversation, but the senior programmer may come up with ideas you wouldn't think of. They may have more innate experience in the domain which might pull you away to very different implementations from what you had in mind initially. This is all good, but a common pitfall with LLMs is that they're much more conformative than a senior engineer, much faster, much more responsive. This is where it gets dangerous: they don't tell you you're wrong 80% of the time, like a rubebr duck. They also have innate experience of a domain, like a senior engineer. The result is that your brain never gets a 'hang' signal, and it doesn't feel the need to reevaluate.
philipportner
> Are Rubber Ducks offloading thinking?
Just that an actual rubber duck doesn’t do anything. You solve the problem you have by talking, and in doing so, thinking, to come up with a solution, an idea, or gain better understanding.
After that you either implement something yourself or have learned something.
With an LLM you offload all of that, the only thing you still do is tell it what the problem is. The agentic duck does the rest and you look at the output.
Even if you have to argue, you argue without having gone through the steps to gain anything yourself.
duxup
Being able to ask the dumb questions but then get focused on something more technical is so nice. Get examples on demand.
It helps me pickup and understand new things pretty fast.
nemesis17
I feel the same empowerment. LLMs allow me to tap into my second-order knowledge.
jhhh
I agree with the exploration aspect of AI. I don't have it create code for me but I have certain longer threads about topics that I think are interesting but would not have approached without it. Previously you would've had to find an entire textbook or class on a subject that may not have been quite what you were looking for if you were just interesting in poking around in a topic to start. Like I have a long running chat about making databases and WALs that I don't think I'd have started if I had to dive fully into a course on writing a database from scratch (if that even exists).
wolvoleo
Same here. It really helps me get deep into things that would have been daunting to start due to the boring research work required. I have heavy ADHD and high intelligence (not to brag, it is often more of a burden than a blessing!) so I can understand things really quickly but I have no patience for painstaking research or planning and documentation. I just want to build, to create and make my vision come to life but I'm really one-sided about that. I don't care about how it fits in the organisation process, the governance or where it should be documented.
And when I need to learn something new I hate doing courses because they're always too slowly paced for me. All courses are paced for the slowest possible participant. Especially those from the US where trainers have this "training voice" where they speak slowly and articulated and usually non-skippable which makes me scream in agony. My brain can't slow down to that level, I just lose focus then. It's like driving a ferrari at 5km/hr. Especially the non-IT courses are so slow and constantly repeat everything, as if I didn't fully understand them the first time, argh. So I tend to spend my time on these awfully designed mandatory HR courses trying to fuck around in developer mode and re-enable the skip segment features :) The more boring the training the more I learn about moodle and its frameworks.
I just love to dive right in and just do something but I often skip that first analysis of what method would be best. And I tend to stick with what I know. Which is often not the best way in a greenfield project. And often the boring first steps are so awful for me that I just put it off. If I do start I often do it half-assed. Then I run into a wall, have to regroup etc. I learn from that (mistakes are always a learing experience), but it's slow going because I take the wrong direction first.
AI really helps me get past that first step meaning I actually start it instead of procrastinating the research forever and move onto the next thing. And I can do great things because I know what I want and how I want it. It's a real enabler for me especially something like cowork. It helps me widen my horizon efficiently instead of wasting my time and energy like online courses do. It picks up things that my weak points are (patience) and feeds me information in a highly dense format I can easily digest. Helped by custom prompts of course ("don't tell me I make great decisions, evaluate everything on its merits, just tell me the meat of the content, don't elaborate"). I still work along with it and take ownership of its output, I'm way too much of a control freak to let the AI do everything without understanding it. But it gets me there quicker and it avoids procrastination by skipping the boring steps. I ask it to suggest methods and techniques, compare their pros and cons so I fully understand what I'm selecting for.
I have to say this was not really the case with early simple question/answer LLMs. The hallucinations were too bad, the searches too limited. The knowledge too hit and miss. It would come up with wrong stuff or overlook the best options. But the new high-context agentic technologies like cowork that really investigate the source material for the purposes I provide, they really are a game changer for me.
For example, I recently asked it about an obscure error I'd been having. Previous simple LLMs would just collapse into a directionless troubleshooting mode, "Did you try...?". Cowork actually went and retrieved all the relevant code from github without being asked to do so, and evaluated what would have happened in the background for that message to appear. Really good to see because it's exactly what I would have done. But I would have taken much more time to do it.
So no, for me it's not making me lose savviness. It's helping me grow it in exactly the way that's efficient for me.
FinnLobsien
I think the difference is that you read the code and get curious about it rather than glancing at the output and immediately telling it what to correct.
_the_inflator
How large are your projects? What kind of?
Just curious. No right or wrong answer possible.
grebc
Is reading example code uncool?
dvt
Have mathematicians become "less savvy" since the invention of the abacus or calculator? The invention of logarithms or analysis? In some ways, the answer is trivially positive. In fact, number crunching was, for most of human history, a key element of being a maths researcher.
However, when we look at the claim with a more critical eye, it becomes obvious that this view is myopic. In fact, current progress is in fact indebted to the invention various tools, which has led us to grand theories and hypotheses: things the Ancient Greeks could only dream about.
Is coding about writing assembly? Is it about writing boilerplate? About writing endless if-then statements? I would argue that it is not: it's about coming up with ideas, building products, finding customers, iterating, producing code that's easy to read, easy to understand, easy to extend.
Coding (at least the kind LLMs do) was never really the hard part, and I'd like to know specific examples of savvyness the author is losing. They've made me more critical of code and able to create better patterns, simply because I can iterate so quickly.
adamddev1
> Have mathematicians become "less savvy" since the invention of the abacus or calculator? The invention of logarithms or analysis?
This is not a fair analogy. Those things are like compilers, languages, and type-checkers. They are deterministic, carefully proven, hand-crafted tools, that allow people to go safely to a higher level of abstraction and still make things in a deterministic way. The jump into the fuzzy world of LLMs is a completely different thing. It's different than the jump from assembly to a higher-level language.
dvt
Not sure how much frontier math you follow, but probabilistic arguments are extremely popular (particularly in, weirdly enough, number theory), especially for fiddling with the bounds on things, or getting closer to full coverage of various theorems.
So this idea that doing math is some kind of fully deterministic & mechanistic process is a bit misleading.
luke5441
Is that kind of argument something like where you count the number of cranks claiming P=NP and the number of cranks claiming P!=NP and then if you get 95% papers claiming P!=NP that one wins?
dvt
What? Not at all. I'm referring to things like discussing distributions of certain things or probabilities of things as we go towards infinity, etc. These are proper papers, not surveys.
luke5441
But if a LLM writes such a paper, how do you know there aren't any issues in it that cause wrong probabilities/distributions?
dvt
You don't, that's why you review it (just how you would handle a real human submitting a paper). The main issue with LLM papers is that they tend to be very dense/circuitous or—because LLMs don't truly understand what they're doing—have the wrong focus. An LLM might spend pages on a trivial result, but quickly gloss over a truly remarkable finding.
Terry Tao talks about this challenge in some of his blog posts and interviews, it's pretty interesting.
luke5441
That was adamddev1s point. It's a unreliable tool. You cannot copy & paste its results into a paper as proof. You could do this with e.g. Lean.
To review what it writes you need to understand the topic to such an extend that you could have written it yourself (at least that is how it is for me when I review code).
Yes, for inspiration or search they are great tools.
coldtea
>You don't, that's why you review it
Review enough of those without doing the proving yourself, and you'll soon lose the ability to review. And have LLMs do the course work for you at math school, and you might never acquire such ability to begin with anyway.
adamddev1
Right, programmers also use randomly generated info from property based testing for example? But I was referencing the analogy to the abacus, calculator, logarithms, analysis, and higher level languages.
coldtea
>Have mathematicians become "less savvy" since the invention of the abacus or calculator?
The abacus and calculator doesn't do high order thinking or prove theorems for you. Math as a field aren't about manually adding numbers together or doing simple algebra formulas, and wheren't that even 2000 years ago.
Those calculators are to math what something like an IDE or a helper library is to programming.
The analogy would be with theorem provers, and even better, with the exact same tool as in the programming case: LLM doing the proofs.
If mathematicians had LLMs and theorem provers do their job, they'd certainly have gotten "less savvy". And if they let them write their coursework, starting from university, they'd might never even become savvy enough in the first place.
_the_inflator
Did you study math? I did and I cannot relate to your comparison.
Mathematics isn’t about calculating numbers but formulas and proves.
I have to put it politely: until the advent of AI no machine ever “calculated” a formula for me, because I didn’t have to calculate anything.
It is like driving a Tesla and laughing about the manufacturer when he mentions how hard it is to get the product right, because you have an easy ride.
I can relate to the article’s AI depression sometimes.
Whatever comes to your mind, AI makes it happen in a couple of minutes in most cases no matter what you imagined.
You invest days and the AI just build it is a couple of minutes way better than you and your 10 teams together did before.
So much hard work and sacrifice - and now? Poof - there it is.
It is nice but also depressing. And not being challenged is the actual problem.
pintert3
I think he means the savviness in the art of programming the ideas. The skill and art involved in finding interesting and efficient ways to effect ideas and solutions to problems. It is(or was) fun for a good number of us who joined before AI changed the game. Now its dismissed tending to a waste of time, hence the supposed loss in savviness over time.
grebc
I’d argue you’ve got a very VC focused view of coding.
zerof1l
Prior to AI, I’d spend hours debugging and figuring how to make something work. It felt exciting in the end; the journey not so much. It was frustrating sometimes, and the overall output was not great. Sometimes I couldn’t try out all of my ideas simply because I had no capacity to do it. Anything requiring multiple months of work, I’d get burned out and never finish anything.
Now I delegate all the grinding to AI. I still feel accomplishment when I get the final result. But more importantly, I can now build something in couple of weeks that would otherwise take me half a year. I can focus now less on the for-loop vs while-loop and more on the big picture, like keeping things tidy and separation of concerns. The front-end and how end-users will interact with what I build is now more important to me, as well as the design. Previously, I had brain capacity only for writing my clever backend algorithm. Now I can make an equally clever front-end for it.
Maybe we never enjoyed the grind? We just came to accept it and convinced ourselves to enjoy it, because there was no other way. If you wanted something cool, it meant you had to put in hours. Now you just need to word it correctly and get the prerequisites.
pintert3
This is exactly my predicament of recent. I haven't necessarily feel like I've lost the savviness yet, but I feel I've lost the fun of the job. I guess it's the reality of life that all fun work eventually has its end of fun. Even the foot messengers of the past now probably load trucks/trains and sit the whole day, no longer feeling the fun of meeting people on the journey through whatever street or town they pass.
skeledrew
Begging the question here. Author is aware that they can choose not use LLMs and they... don't? It's like living at walking distance from town, being used to going in and around on foot, then electric scooters became available, so they start using them to get around while complaining that they aren't getting the exercise they want anymore.
aogaili
While I also griefed the years spent mastering craft being automated.
I also had a lot of ideas that I wanted to execute and I'm finally able to do so. Some of them are wild and big, unthinkable last year. I also really enjoy building product/software that people are delighted to use. So from my perspective, the net positive outweighs the losses.
My work position is less certain, I grief the years spent mastering things just to see AI doing it, but I went to software engineering because I enjoyed building products, from ideas to something people can play with, and this has never been easier.
newsomix9xl
Using AI can open up topics you couldn't work on before.
I think you're blaming tools for what are other problems.
No carpenter ever said "I used to plane wood by hand and carve magnificent pieces, now that I have an orbital sander and power tools, the love just isn't there".
I don't buy this argument at all.
klez
It's like every time we discuss this on hn we have to go back to basics.
Builders/makers can be roughly subdivided into two camps: some like building for the final products; others like building for the process itself.
I'm in the latter camp. OP probably too. You probably are in the former.
And that's fine. But let's just stop pretending the other side doesn't exist.
EDIT: And this goes for OP as well. They forgot a lot of "for me"s in the article.
thot_experiment
You don't need the "for me", it's implicit in it being an article written by someone in a casual setting. I'd argue the issue is that we as internet blog post readers treat these think pieces with more gravitas then they deserve as a baseline.
molly0
Good point
croo
The example is not helpful though, probably a lot of carpenter said similar things. knowledge and attention to detail is absolutely lost with power tools and loud machines can take away a lot from the experience of working with wood. It also made the work more dangerous by instantly removing fingers and inhaling powder that was just not there before. It gave carpenters speed in return. Can you maybe give a better example?
furyofantares
> No carpenter ever said "I used to plane wood by hand and carve magnificent pieces, now that I have an orbital sander and power tools, the love just isn't there".
I would guess probably lots did say that?
At the very least I have no idea why you would be able to claim otherwise, we are far removed in time from when there were carpenters straddling the introduction of power tools.
I think more apt analogy is that now I get random parts from Ikea and screw them together to something that resembles furniture. Just need to get enough parts and eventually they mostly fit. Customers really don't notice any difference.
aninteger
The only problem with this analogy is that LLMs are not orbital sanders... Or if they are, they are orbital sanders that operate correctly only 95% of the time. The other 5% of the time they do the wrong thing and cause you to do more work and/or damage the wood (not a great analogy though).
goodmythical
>they are orbital sanders that operate correctly only 95% of the time. The other 5% of the time they do the wrong thing and cause you to do more work and/or damage the wood
So more of a home-gamer chainsaw? Except that I'm not positive the once/twice per year chainsaw is nearly so effective.
anon7000
Power tools aren’t exactly the paragon of determinism. It’s ways to make mistakes with them, easy for wood to get caught the wrong way, relatively easy to injure yourself.
xnzakg
not a bad analogy imo. power tools vs hand tools. faster, but if you're not careful you can cause more damage before you notice. like a table saw, where you can cut too far or at an angle and the result will look bad. or an electric screwdriver, where you don't feel the torque as much, so it's easier to cross-thread.
and skills from building with hand tools help you judge the results, but not necessarily with actually using the tools.
mysterydip
But that’s not the right analogy, in my opinion. It would be a carpenter saying “I used to carve by hand, but now I tell the carving machine to make a chair and it spits it out.”
titularcomment
Yes, using analogies of comparing LLMs to relatively simple tools and on the same plane declaring them the predecessors to AGI (or for some, we have achieved AGI) is a common hypocritical take.
anon7000
Maybe you’re partly right, maybe not, but I don’t think this is a good analogy. Let’s say output is the most important thing, which is the implication here. But why do I care about my output? Why does a woodworker care about their output?
For a woodworker, I think your analogy should be a “furniture factory,” not power tools. (Power tools in software would be like a great IDE and powerful language and framework tooling; AI is… way more than that.) I bet a “woodworker” in a furniture factory (if that even exists) doesn’t much care about their output, and doesn’t love the process anymore.
And getting paid for it doesn’t make you care about your work either, it just makes you care about money you can use on other things you do care about.
It would be foolish for anyone looking at the great change happening in software, and ignore the fact that we’re crossing a line where it’s less about craft, and where developers are just managing factory-like processes. Which is very different from what most signed up for.
And of course you can still do your craft, just like there are custom woodworkers as well as factories. But the space for that in AI-driven companies is vanishing as much as we might hate it or push back against it.
And of course there are also people like you who like the opportunity that more output provides, or who use it more like a tool and less like a factory. And that’s fine.
But no, the transition happening in the industry is not deliberately trying to keep the craft of it or even to keep developers interested in the job.
The transition is turning people who input stuff into a software factory and do QA on the output, or turning people into “factory automation engineers.” Those are both very different jobs from what people who like programming signed up for.
And yeah maybe no one “deserves” to have craft in their jobs. But let’s be honest, it is very natural, and very human, to be upset when the main motivation and reason for doing your work disappears. The same thing has happened in countless industries, and people have been talking about it for decades if not hundreds of years.
E.g. small-time farming replaced by exploitative big agriculture. A farmer might have a reason and identify in the work before, owning their own bit of land they tender. And now if they’re just hired to drive a tractor, they don’t. It’s a massive difference, and no wonder our society is increasingly depressed.
thot_experiment
What are you talking about lmao, tons of carpenters have said that. The subjective experience of your craft is shaped by the tools you chose to use. For some people, and I venture this is most often true for master craftspeople, not only is the journey is more important than the destination, but the limitations of the path breed the creativity needed to even arrive at the destination.
cliche
> I’ve experienced that great feeling of “this is a great idea” coming out of nowhere while doing unrelated activities, and having to rush back at my desk to write it down and implement it. That was great - it was rewarding, it was really meaningful.
This is it for me. It’s a pretty incredible tool but it killed the vibe and challenge of building things.
A_D_E_P_T
It's true that LLMs have trivialized some aspects of software development. (We're at the "software writes itself following natural language inputs" stage.)
What's nevertheless true is that they've hardly touched hardware hacking. There are so many interesting opportunities right now in robots, drones, materials, batteries, and custom embedded systems that still demand dirty hands and physical intuition. These still reward savvy, and they're fun!
dumberquestions
I remember feeling a little bad the first time seeing AI create a UI as good as the one I did using a niche library and only a picture reference, when I focused on the feeling I realized it was an ego issue, nothing is stopping me from doing it by hand if I enjoyed it for itself, it's just that what I did was no longer scarce or difficult.
ngruhn
Not sure it's an ego issue for everyone. If your job was fun and now it's not anymore, you're having 8h a day less fun.
wdm0006
The trick is to make, and care deeply about, fewer bigger decisions. Savviness will always matter but the things to be savvy about change.
MattPalmer1086
I am finding the opposite, but then I mostly work on algorithms and writing a lot of code to test various approaches to improve them is boring.
Today I had claude figure out that using SIMD would not improve performance, and it gave me an analysis of why not. So my savvy improved without having to spend days figuring that out.
Having said that, I get that using AI can also make things less fun and reduce your learning too. I guess it depends on what you use it for.
ChuckMcM
From the article: With the LLMs every time there’s a mistake, I don’t learn. I can spot the mistake because I have some experience (lucky enough to have been born 34 years ago), I can pinpoint the mistake and let the LLM fix it.
This is something I realized when I became a manager of programmers rather than doing the work myself. When I write code I do it from a position of having a deep understanding of what all the moving pieces are, when someone working for me writes code I just have to trust they had that level of understanding[1].
But being a manger, the thrill of solving the puzzle was replaced by the thrill of helping some new engineer get better at the craft of writing code. They took pride in their growing skill set and I took pride in helping them grow.
Managing consultants however was very transactional. Did they deliver what was asked of them? Did they fix the flaws I pointed out? Do I still want to pay them?[2] And I could of course fire them easily and try new ones that had a reputation for being "better" at the tasks I needed done.
If history is any judge, there have been many companies over the years where some founder/leader has a "great idea" and just needed someone technical to make it real. And the typical arc for those companies was they get founded, they spend a ton of money on consultants to build their technology, something changes (fashion, systems, something) and their product is no longer a good fit for the market so they try to hire consultants to fix it, which now costs more than 2x because first the consultants had to understand what the previous consultants did, then they have to change it in a way that leadership wants. The result is a product that got more and more brittle until there wasn't enough money to or enough consultants to anneal the product back into something people would pay enough for to support the salaries of the visionary leaders who never had any idea at all of how the product they sold actually worked.
It feels like we're going to speed run that. In the LLM era.
[1] And of course that level of trust is not always warranted but that's a different comment :-)
[2] With consultants there is a velocity vs organizational expertise equation that always needs to be balanced. Dropping people into a legacy code base they've never seen is just as hard for them as dropping them into a consultant's work product that they hadn't previously seen.
jkahrs595
Just stop using them ffs. So sick of this “we lost the craft” nonsense. Just do it the way you like doing it.
adammarples
That's not how addictions work
system2
This article tells me the author is not that creative. With LLMs, the number of projects I create increases by 1000%.
geophph
Being creative and creating things are not the same?
system2
Isn't being creative a requirement to create things? The author complains that he can't do the basic things that LLMs now take care of. And I am saying that if you really have a creative mind, you can come up with 1000s of projects every day and enjoy bringing them to life very fast.
For example, I wanted to build my own note-taking app for a long time because I got sick of Obsidian. Now I built my own, and it is magnitudes better than Obsidian, at least to me and for my use. I built smart home systems with Claude because, without an LLM, it would be too time-consuming to troubleshoot APIs and read all the api documentation. I wanted to test a TTS model, and it took me 10 minutes waiting to test it without reading the documentation. I use it for everyday random creations all day long.
Instead of one tiny project consuming all my week, I can now work on all of them in hours. My /srv folder has over 100 projects of my own, and some of them are parent folders to smaller dependency projects.
phoghed
I keep trying to get the product people I work with to do personal vibe coding and they haven’t yet. Feels like they should have more ideas than I do.
aogaili
I think that is the real bottleneck now. Creative people used to be hindered by execution, now you can aim higher.
Of course there are errors and gaffes and weird bits all over, but that helps keep me in the loop and understanding it better.
I can now (more) boldly go where I feared to tread before.
To be clear, I am closing in on 60 years old. I’ve written a lot of code in a lot of languages. I have not bothered to check every fad that has come down in front end web and Javascript for the past umpteen years. Front end hasn’t been my thing. With current AI, I can leap frog a lot of bull crap I don’t care about and see what at least the AI thinks is good. I still retain a brain, I can still investigate what it did and form an opinion about it.
I treat the AI stuff as idiot savants that can make really nice stuff or go totally off in the weeds. But it is more than good enough to help me be a better me at what I do.
Uh, have you asked it?
These sorts of articles are funny. Like the author assumed that you have to use an LLM in a certain way.
“I can just ask the LLM and there it is! I didn’t learn anything! :(“
You can interrogate the LLM. You can ask it all sorts of things. Tell it to justify its decisions. Tell it to give you supporting evidence. Tell it to give you options. Tell it to explain things in detail.
I’ve been using it as a learning tool and it’s quite effective.
Granted the stuff I don’t care about really does get worse. If I have a config typo I just go “There’s a typo in blah. Find it and fix it.” Or setting up the boiler plate for a project. I’m getting quite useless at that.
But it doesn’t have to be any particular way. It’s a tool. Exploit it.
Writing things down and then arguing with an agent that constantly gets things wrong is just a new kind of Rubber Duck.
Are Rubber Ducks offloading thinking?
These are more like search plus memory tools.
Memory loss is real but that has been happening for me way before agents.
Depends on your usage. A rubber duck won't upheave the entire history of written human knowledge to reach for your answer. It doesn't have inherent bias, and its purpose could also be handled by a wall, or the air infront of you if you can stretch your mind a little. What I'm going for here is that as you said, its just a hack to make your brain look for answers in a more efficient manner, by clearly underlining your thought process out loud, tricking yourself into conversation mode, yada yada. Its still you doing all the work, you're just being smarter about it.
A LLM is more akin to a senior pair programmer. It may seem like a rubber duck on the surface, in the sense that you're having a conversation, but the senior programmer may come up with ideas you wouldn't think of. They may have more innate experience in the domain which might pull you away to very different implementations from what you had in mind initially. This is all good, but a common pitfall with LLMs is that they're much more conformative than a senior engineer, much faster, much more responsive. This is where it gets dangerous: they don't tell you you're wrong 80% of the time, like a rubebr duck. They also have innate experience of a domain, like a senior engineer. The result is that your brain never gets a 'hang' signal, and it doesn't feel the need to reevaluate.
Just that an actual rubber duck doesn’t do anything. You solve the problem you have by talking, and in doing so, thinking, to come up with a solution, an idea, or gain better understanding.
After that you either implement something yourself or have learned something.
With an LLM you offload all of that, the only thing you still do is tell it what the problem is. The agentic duck does the rest and you look at the output.
Even if you have to argue, you argue without having gone through the steps to gain anything yourself.
It helps me pickup and understand new things pretty fast.
And when I need to learn something new I hate doing courses because they're always too slowly paced for me. All courses are paced for the slowest possible participant. Especially those from the US where trainers have this "training voice" where they speak slowly and articulated and usually non-skippable which makes me scream in agony. My brain can't slow down to that level, I just lose focus then. It's like driving a ferrari at 5km/hr. Especially the non-IT courses are so slow and constantly repeat everything, as if I didn't fully understand them the first time, argh. So I tend to spend my time on these awfully designed mandatory HR courses trying to fuck around in developer mode and re-enable the skip segment features :) The more boring the training the more I learn about moodle and its frameworks.
I just love to dive right in and just do something but I often skip that first analysis of what method would be best. And I tend to stick with what I know. Which is often not the best way in a greenfield project. And often the boring first steps are so awful for me that I just put it off. If I do start I often do it half-assed. Then I run into a wall, have to regroup etc. I learn from that (mistakes are always a learing experience), but it's slow going because I take the wrong direction first.
AI really helps me get past that first step meaning I actually start it instead of procrastinating the research forever and move onto the next thing. And I can do great things because I know what I want and how I want it. It's a real enabler for me especially something like cowork. It helps me widen my horizon efficiently instead of wasting my time and energy like online courses do. It picks up things that my weak points are (patience) and feeds me information in a highly dense format I can easily digest. Helped by custom prompts of course ("don't tell me I make great decisions, evaluate everything on its merits, just tell me the meat of the content, don't elaborate"). I still work along with it and take ownership of its output, I'm way too much of a control freak to let the AI do everything without understanding it. But it gets me there quicker and it avoids procrastination by skipping the boring steps. I ask it to suggest methods and techniques, compare their pros and cons so I fully understand what I'm selecting for.
I have to say this was not really the case with early simple question/answer LLMs. The hallucinations were too bad, the searches too limited. The knowledge too hit and miss. It would come up with wrong stuff or overlook the best options. But the new high-context agentic technologies like cowork that really investigate the source material for the purposes I provide, they really are a game changer for me.
For example, I recently asked it about an obscure error I'd been having. Previous simple LLMs would just collapse into a directionless troubleshooting mode, "Did you try...?". Cowork actually went and retrieved all the relevant code from github without being asked to do so, and evaluated what would have happened in the background for that message to appear. Really good to see because it's exactly what I would have done. But I would have taken much more time to do it.
So no, for me it's not making me lose savviness. It's helping me grow it in exactly the way that's efficient for me.
Just curious. No right or wrong answer possible.
However, when we look at the claim with a more critical eye, it becomes obvious that this view is myopic. In fact, current progress is in fact indebted to the invention various tools, which has led us to grand theories and hypotheses: things the Ancient Greeks could only dream about.
Is coding about writing assembly? Is it about writing boilerplate? About writing endless if-then statements? I would argue that it is not: it's about coming up with ideas, building products, finding customers, iterating, producing code that's easy to read, easy to understand, easy to extend.
Coding (at least the kind LLMs do) was never really the hard part, and I'd like to know specific examples of savvyness the author is losing. They've made me more critical of code and able to create better patterns, simply because I can iterate so quickly.
This is not a fair analogy. Those things are like compilers, languages, and type-checkers. They are deterministic, carefully proven, hand-crafted tools, that allow people to go safely to a higher level of abstraction and still make things in a deterministic way. The jump into the fuzzy world of LLMs is a completely different thing. It's different than the jump from assembly to a higher-level language.
So this idea that doing math is some kind of fully deterministic & mechanistic process is a bit misleading.
Terry Tao talks about this challenge in some of his blog posts and interviews, it's pretty interesting.
Yes, for inspiration or search they are great tools.
Review enough of those without doing the proving yourself, and you'll soon lose the ability to review. And have LLMs do the course work for you at math school, and you might never acquire such ability to begin with anyway.
The abacus and calculator doesn't do high order thinking or prove theorems for you. Math as a field aren't about manually adding numbers together or doing simple algebra formulas, and wheren't that even 2000 years ago.
Those calculators are to math what something like an IDE or a helper library is to programming.
The analogy would be with theorem provers, and even better, with the exact same tool as in the programming case: LLM doing the proofs.
If mathematicians had LLMs and theorem provers do their job, they'd certainly have gotten "less savvy". And if they let them write their coursework, starting from university, they'd might never even become savvy enough in the first place.
Mathematics isn’t about calculating numbers but formulas and proves.
I have to put it politely: until the advent of AI no machine ever “calculated” a formula for me, because I didn’t have to calculate anything.
It is like driving a Tesla and laughing about the manufacturer when he mentions how hard it is to get the product right, because you have an easy ride.
I can relate to the article’s AI depression sometimes.
Whatever comes to your mind, AI makes it happen in a couple of minutes in most cases no matter what you imagined.
You invest days and the AI just build it is a couple of minutes way better than you and your 10 teams together did before.
So much hard work and sacrifice - and now? Poof - there it is.
It is nice but also depressing. And not being challenged is the actual problem.
Now I delegate all the grinding to AI. I still feel accomplishment when I get the final result. But more importantly, I can now build something in couple of weeks that would otherwise take me half a year. I can focus now less on the for-loop vs while-loop and more on the big picture, like keeping things tidy and separation of concerns. The front-end and how end-users will interact with what I build is now more important to me, as well as the design. Previously, I had brain capacity only for writing my clever backend algorithm. Now I can make an equally clever front-end for it.
Maybe we never enjoyed the grind? We just came to accept it and convinced ourselves to enjoy it, because there was no other way. If you wanted something cool, it meant you had to put in hours. Now you just need to word it correctly and get the prerequisites.
I also had a lot of ideas that I wanted to execute and I'm finally able to do so. Some of them are wild and big, unthinkable last year. I also really enjoy building product/software that people are delighted to use. So from my perspective, the net positive outweighs the losses.
My work position is less certain, I grief the years spent mastering things just to see AI doing it, but I went to software engineering because I enjoyed building products, from ideas to something people can play with, and this has never been easier.
I think you're blaming tools for what are other problems.
No carpenter ever said "I used to plane wood by hand and carve magnificent pieces, now that I have an orbital sander and power tools, the love just isn't there".
I don't buy this argument at all.
Builders/makers can be roughly subdivided into two camps: some like building for the final products; others like building for the process itself.
I'm in the latter camp. OP probably too. You probably are in the former.
And that's fine. But let's just stop pretending the other side doesn't exist.
EDIT: And this goes for OP as well. They forgot a lot of "for me"s in the article.
I would guess probably lots did say that?
At the very least I have no idea why you would be able to claim otherwise, we are far removed in time from when there were carpenters straddling the introduction of power tools.
So more of a home-gamer chainsaw? Except that I'm not positive the once/twice per year chainsaw is nearly so effective.
and skills from building with hand tools help you judge the results, but not necessarily with actually using the tools.
For a woodworker, I think your analogy should be a “furniture factory,” not power tools. (Power tools in software would be like a great IDE and powerful language and framework tooling; AI is… way more than that.) I bet a “woodworker” in a furniture factory (if that even exists) doesn’t much care about their output, and doesn’t love the process anymore.
And getting paid for it doesn’t make you care about your work either, it just makes you care about money you can use on other things you do care about.
It would be foolish for anyone looking at the great change happening in software, and ignore the fact that we’re crossing a line where it’s less about craft, and where developers are just managing factory-like processes. Which is very different from what most signed up for.
And of course you can still do your craft, just like there are custom woodworkers as well as factories. But the space for that in AI-driven companies is vanishing as much as we might hate it or push back against it.
And of course there are also people like you who like the opportunity that more output provides, or who use it more like a tool and less like a factory. And that’s fine.
But no, the transition happening in the industry is not deliberately trying to keep the craft of it or even to keep developers interested in the job.
The transition is turning people who input stuff into a software factory and do QA on the output, or turning people into “factory automation engineers.” Those are both very different jobs from what people who like programming signed up for.
And yeah maybe no one “deserves” to have craft in their jobs. But let’s be honest, it is very natural, and very human, to be upset when the main motivation and reason for doing your work disappears. The same thing has happened in countless industries, and people have been talking about it for decades if not hundreds of years.
E.g. small-time farming replaced by exploitative big agriculture. A farmer might have a reason and identify in the work before, owning their own bit of land they tender. And now if they’re just hired to drive a tractor, they don’t. It’s a massive difference, and no wonder our society is increasingly depressed.
This is it for me. It’s a pretty incredible tool but it killed the vibe and challenge of building things.
What's nevertheless true is that they've hardly touched hardware hacking. There are so many interesting opportunities right now in robots, drones, materials, batteries, and custom embedded systems that still demand dirty hands and physical intuition. These still reward savvy, and they're fun!
Today I had claude figure out that using SIMD would not improve performance, and it gave me an analysis of why not. So my savvy improved without having to spend days figuring that out.
Having said that, I get that using AI can also make things less fun and reduce your learning too. I guess it depends on what you use it for.
This is something I realized when I became a manager of programmers rather than doing the work myself. When I write code I do it from a position of having a deep understanding of what all the moving pieces are, when someone working for me writes code I just have to trust they had that level of understanding[1].
But being a manger, the thrill of solving the puzzle was replaced by the thrill of helping some new engineer get better at the craft of writing code. They took pride in their growing skill set and I took pride in helping them grow.
Managing consultants however was very transactional. Did they deliver what was asked of them? Did they fix the flaws I pointed out? Do I still want to pay them?[2] And I could of course fire them easily and try new ones that had a reputation for being "better" at the tasks I needed done.
If history is any judge, there have been many companies over the years where some founder/leader has a "great idea" and just needed someone technical to make it real. And the typical arc for those companies was they get founded, they spend a ton of money on consultants to build their technology, something changes (fashion, systems, something) and their product is no longer a good fit for the market so they try to hire consultants to fix it, which now costs more than 2x because first the consultants had to understand what the previous consultants did, then they have to change it in a way that leadership wants. The result is a product that got more and more brittle until there wasn't enough money to or enough consultants to anneal the product back into something people would pay enough for to support the salaries of the visionary leaders who never had any idea at all of how the product they sold actually worked.
It feels like we're going to speed run that. In the LLM era.
[1] And of course that level of trust is not always warranted but that's a different comment :-)
[2] With consultants there is a velocity vs organizational expertise equation that always needs to be balanced. Dropping people into a legacy code base they've never seen is just as hard for them as dropping them into a consultant's work product that they hadn't previously seen.
For example, I wanted to build my own note-taking app for a long time because I got sick of Obsidian. Now I built my own, and it is magnitudes better than Obsidian, at least to me and for my use. I built smart home systems with Claude because, without an LLM, it would be too time-consuming to troubleshoot APIs and read all the api documentation. I wanted to test a TTS model, and it took me 10 minutes waiting to test it without reading the documentation. I use it for everyday random creations all day long.
Instead of one tiny project consuming all my week, I can now work on all of them in hours. My /srv folder has over 100 projects of my own, and some of them are parent folders to smaller dependency projects.