Lots of mentions of the term mourning... As they say in my country, don't sell the skin until you kill the bear.
All I'm seeing around me is people dropping best practices in a FOMO driven push for speed: let's stop reviews, let's drive 5 agents in parallel, let's not even look at the code!
This is going to blow up.
Only after we pick up the remains we'll find a more sustainable approach for AI usage. I suspect that version will still require crafters.
If we end up in a place where the craft truly is dead, then congratulations, your value probably just dropped to zero. Everyone who's been around startup culture knows the running jokes about those 'I have a great idea, I just need someone to code it' guys. Now you're one, and you'll find how much ideas are worth.
> If we end up in a place where the craft truly is dead, then congratulations, your value probably just dropped to zero
I think, then that the value of all knowledge work will have dropped to zero. Software engineering is, to my mind, “intelligence complete.” If you can do it with knowledge work, you can have software do it.
That’s not the point of nor the reason for knowledge work.
The fundamental mechanism of knowledge work is people. They haven’t changed at all. And what they need to understand and learn hasn’t changed. All the agents in the world and all of the methane guzzling data centres can’t tell you what to write in the specification nor if what the computer has generated faithfully implements that specification.
Yep. Most knowledge work is about coordination between people and transporting the right information to the people that are thinking and the people that are doing (not always separate groups, but can be one group switching mode). You need people because there's a lot of shared context between individuals in society that is not encoded anywhere.
A year ago everyone was sure these things couldn’t write functional code. A few months ago people started saying they need to be operated by people who could otherwise write the code.
Like… the guy in Office Space who took specifications from the customer via a secretary and gave those to the engineers (and we know what happened to him).
But I’m not sure that’s a thing, at least for long, either. The original super power of these things wasn’t that they could write code. It was that they could very competently extract meaning from natural language, debug what you were saying from the terrible way you expressed it, and still formulate competent answers.
That doesn’t sound like a comfortable place for former devs to sit for the next few years.
I have an idea for an “evidence editor.” Claude is waiting for me to tell it exactly what I want this thing to be. But I don’t know. I haven’t figured out how to square the various circles, even in my fantasies. Until I do, Claude sits and waits. And waits…
A couple of guys at work have been raving about Claude. How quick they get stuff done, how great the code is, how working any other way is a waste of time.
I just had the misfortune today to wade into one of their codebases. It's 60k lines of code for something that should have been simple, and it's an absolute fucking mess. I'm gonna have to rip out most of it and start over just to get it to do what we actually need it to do.
I use LLMs, they come in handy, and I use agents, but this "have agents do everything" nonsense is a disaster, and it's only going to get worse.
On the upside I'm getting paid to fix this shit show.
> On the upside I'm getting paid to fix this shit show.
A lot of my career has been this, not due to choice but circumstance. Startups write terrible code, in general. Enterprises write terrible code. I’ve worked with both. If it becomes important enough, someone has to fix it at some point.
Current AI models seem to be job security machines for that kind of work.
We are way past wringing our hands over agentic engineering. Every startup and all fast moving companies are onboard. They don't hand code anymore. There will not be some code quality crisis that will stop everyone in their tracks. I'm trying to cope with this too, but I don't think the best path is praying for failure.
Just out of the popularity of the claim, I’ll bite.
Both big tech and startups are now full of people working at 10x, features are written as fast as PMs can think them, monoliths self heal with agents buzzing over them.
10x means 10 times the outcomes in a given amount of time, so did you see the last iOS version pack a decade worth of features in a single release?
Do you remember when meta moved their backend to rust in a month?
What about Microsoft software not having a single bug in a year?
At 11 this morning, I wanted to both debug an issue and take a meeting before lunch. Before AI, I would have had to just start debugging after lunch, there wouldn't have been enough time to do both. But now I had Claude debug the issue concurrently with the meeting. Its answer didn't actually make sense (I still do think I'm smarter than Claude, although the gap is narrowing!), but it showed enough of its work that I could make a good guess about what was really going on, and when I asked it to check my hypothesis I got back from lunch with some debug logs that confirmed I'd found the bug.
I can easily believe that, I agree Claude has applications.
I am disputing the idea that this is enough of a game changer to make us mourn our now lost craft. Also, I’m mentioning that we’ve discovered a world of footguns dressed as shortcuts, which we’re not taking proper care of.
First, your experience was required for that story to have a happy ending. Second, we both know someone else could probably have gone with Claude’s senseless hypothesis, asked for a fix and sent it for review. This last part is becoming pretty universal.
10x is definitely possible at a startup level. I suppose not in a big tech world (seems obvious to me, and it's not like development speed was the bottleneck there either, right?)
You can choose not to believe what I say (and I genuinely understand if you do), and I can simply keep on doing it. I'm not taking it out of thin air either. tuesday I did in 8h the work I scheduled for roughly 65h. Ok, so maybe it's not 10x, maybe it's 8x, same ball park.
And that's only talking about development. If I now get into other aspects....I have just spent the last 1.5 hours creating an incredibly detailed backcatalog of tasks and epics. This is the most detailed I have ever done so in my life and it has been working very well. It's like we merged the good of highly-detailed waterfall with the speed of agile.
Tee-hee: Watergile. (I'm sure some expert in the field will let me know I have coined a new term for something that very much has a name; excuse my ignorance in advance).
Nonetheless, I did this all by talking to the computer which is interfacing with my project management tools, the project documentation, and the project code. Full context on everything. In the past, I would have taken 3 or 4 days to create the same amount of tasks with a vaguely similar amount of detail. But, in truth, I wouldn't have spent so much time putting this love into the craft (!!) of planning a project, because it would exhaust me and feel like a waste of time.
Don't get me wrong, I totally see shit code being thrown everywhere by inferior AI models or people who can't tame the beasts, but the right people in my life are _clearly_ building out more, better tested code, and actually built with more care. Maybe it's not at the line-by-line level, but it certainly is from the end-product result (thinking of the actual end-user). I accept your mileage may vary -- this is my very personal experience.
Maybe it'll stop happening, who knows. Maybe price will be prohibitive, or maybe we'll have such an avalanche of ideas that weren't worth building that everyone will be overwhelmed and take a step back. Or maybe we won't develop juniors into the seniors of tomorrow. Or maybe everything will indeed implode once products are large enough that the original development speed can't be maintained anymore and expectations are mismatched.
What I do know is that it is definitely happening in my world, and I haven't had this much fun since I was a little kid learning to code.
That's a very good question. Some metrics I'm watching or considering watching right now:
- Amount of leads we're taking in (per unit of time)
- Sprint velocity changes (task complexity should stay roughly the same with AI, and team velocity increase — we've been seeing this happening)
- Hire rates (more sales people, less developers?)
- Number of projects per unit of time (of similar dimension, hard to measure)
- Length of "bugfixing buffer" before big releases (we've actually been noticing this go down)
- Another way of saying it is: number of bugs, or bugs per feature
- Drift between planned execution time and actual execution time (we've been delivering early...but I guess we'll soon adjust our estimates...or maybe not, who knows?)
- Spend on AI models
- I can't measure this, but I can sort of "feel it": but the overall feedback we get from clients, the feeling we get from them.
- Number of tests (tests have skyrocketed. Can't be sure about the quality, but, hey, it's a metric)
- Feature turnaround time (how long since "feature is proposed" until it's actually implemented)
- documentation to code ratio (not sure what we'll make of it, but there's a somewhat worrying trend here)
- team balance: is everyone slowly becoming fullstack? Do we feel that those who aren't are significantly affecting development speed? if so, that indicates that the other ones are somehow moving faster
I can't really think of any others, but I'm sure they exist.
The future will not be evenly distributed. You can't expect to see it in the productivity of the industry as a whole, or even the productivity of a large company. You might be able to see it in a medium-sized team if you measure carefully.
I think I agree with you. I didn't mean to imply that big corps will build faster -- in fact I said I don't expect them to right now, and I'm not so sure that will change.
What I believe is that early prototype development and pivoting is insanely fast now. And if you find excellent engineers who are also great product people, and then pair them with people who have truly great ideas, many wonderful new products will emerge.
So, it sounds like measuring directly in individual individuals or companies might be tough. (Unless the company is medium-sized?) Maybe we could look for broader trends in the economy and beyond. What sorts of companies will this manifest in? I.e. mostly "tech" companies, or beyond?
> I don't think the best path is praying for failure. Embrace it
"Embrace failure" is exactly the attitude every company is going to take. They've already been working at bottoming out our expectations.
We should have been running companies out of business with regulations and abandonment when their human-written software leaked our private data to criminals, or when their untested forced-updates shut down our systems and sent our IT teams scrambling, or when their unoptimized code forced us to upgrade our hardware or negated any performance gains we should have seen from investing in upgrades.
The quality, reliability, and security of the software we all use and depend on is going to nosedive, and companies already know they can get away with it. They aren't going to start caring about how we feel about that now. "Pay more and settle for less" is where we are today. "Embrace Failure" is the future we're sprinting towards.
That's absolutely not true. The places that have embraced "agentic engineering" are mostly garbage factories, and lots of places, including plenty of startups and fast-moving companies are staying off of this trend. I recognize that most of the people on this site are just trying to self-promote for their own gig, but the level of misinformation is sometimes just staggering.
I work at a massive health care company. They're 100% on the AI bandwagon and are putting AI everywhere they can. Billing, Software, DevOps, everywhere. If you think you can give an Agent some information and have go to work for some user, its 100% on the table for the company to do and either a) then outsource the rest offshore or b) lay the person off or shrink the department to increase the bottom line.
Your healthcare, right now, is being offloaded to AI agents and bots and this is only the beginning.
I literally just sat through the annual “choose your healthcare” plan bullshit and the “meeting” was literally one of the Hr people pulling up a power point narrated by “AI”. You could tell in the first ten seconds.
You’d think our plans would be cheaper given they’re offloading all this work to agents they don’t have to pay a salary to…right?
It is not just startups or small companies embracing agentic engineering… Stripe published blog posts about their autonomous coding agents. Amazon is blowing up production because they gave their agents access to prod. Google and Microsoft develop their own agentic engineering tools. It’s not just tech companies either, massive companies are frequently announcing their partnerships with OpenAI or Anthropic.
You can’t just pretend it’s startups doing all the agentic engineering. They’re just the ones pushing the boundaries on best practices the most aggressively.
Outwards communication and inside results tend to differ vastly. I’ve heard some true horror stories already from companies who claim they’re doing amazing things with great results.
You should be especially on guard if it’s a publicly traded company, selling AI usage is necessary to appease the market (and thereby C-level stock value).
I was still skeptical at the start of this year, but there seems to be a shift underway. Found the StrongDM Dark Factory docs in Feb and they've netted novel results that have been inspiring enough to keep studying and practicing.
> Before AI, both camps were doing the same thing every day. Writing code by hand. Using the same editors, the same languages, the same pull request workflows. The craft-lovers and the make-it-go people sat next to each other, shipped the same products, looked indistinguishable. The motivation behind the work was invisible because the process was identical.
Helps explain why some people are delighted to have AI write code for them while others are unhappy that the part they enjoyed so much has been greatly reduced.
> That feeling of loss though can be hard to understand emotionally for people my age who entered tech because we were addicted to feeling of agency it gave us. The web was objectively awful as a technology, and genuinely amazing, and nobody got into it because programming in Perl was somehow aesthetically delightful.
I think the real divide is over quality and standards.
We all have different thresholds for what is acceptable, and our roles as engineers typically reflect that preference. I can grind on a single piece of code for hours, iterating over and over until I like the way it works, the parameter names, etc.
Other people do not see the value in that whatsoever, and something that works is good enough. We both are valuable in different ways.
Also, theres the pace of advancement of the models. Many people formed their opinions last year, and the landscape has changed a lot. There’s also some effort requires in honing your skill using them. The “default” output is average quality, but with some coaxing higher quality output is easily attained.
I’m happy people are skeptical though, there are a lot of things that do require deep thought, connecting ideas in new ways, etc., and LLMs aren’t good at that in my experience.
> I think the real divide is over quality and standards.
I think there are multiple dimensions that people fall on regarding the issue and it's leading to a divide based on where everyone falls on those dimensions.
Quality and standards are probably in there but I think risk-tolerance/aversion could be behind some how you look at quality and standards. If you're high on risk-taking, you might be more likely to forego verifying all LLM-generated code, whereas if you're very risk-averse, you're going to want to go over every line of code to make sure it works just right for fear of anything blowing up.
Desire for control is probably related, too. If you desire more control in how something is achieved, you probably aren't going to like a machine doing a lot of the thinking for you.
I think this is a false dichotomy because which approach is acceptable depends heavily on context, and good engineers recognize this and are capable of adapting.
Sometimes you need something to be extremely robust and fool-proof, and iterating for hours/days/weeks and even months might make sense. Things that are related to security or money are good examples.
Other times, it's much more preferable to put something in front of users that works so that they start getting value from it quickly and provide feedback that can inform the iterative improvements.
And sometimes you don't need to iterate at all. Good enough is good enough. Ship it and forget about it.
I don't buy that AI users favor any particular approach. You can use AI to ship fast, or you can use it to test, critique, refactor and optimize your code to hell and back until it meets the required quality and standards.
> nobody got into it because programming in Perl was somehow aesthetically delightful.
To this day I remember being delighted by Perl on a regular basis. I wasn't concerned with the aesthetics of it, though I was aware it was considered inscrutable and that I could read & write it filled me with pride. So yea, programming Perl was delightful.
I strongly disagree. There's always been two camps ... on everything!
Emacs vs. vi. Command-line editor vs. IDE. IntelliJ vs. VS Code. I could do like twenty more of these: dev teams have always split on technology choices.
But, all of those were rational separations. Emacs and vi, IntelliJ and VS Code ... they're all viable options, so they boil down to subjective preference. By definition, anything subjective will vary between different humans.
What makes AI different to me is the fear. Nobody decided not to use emacs because they were afraid it was going to take their job ... but a huge portion of the anti-AI crowd is motivated by irrational fear, related to that concern.
It doesn’t help that the CEOs of these companies are hyping up the fear. It’s no wonder people are afraid when the people making the products are spouting prophecies of doom.
A huge portion of the pro-AI crowd is motivated by irrational hype and delusion.
LLMs are not a tool like an editor or an IDE, they make up code in an unpredictable way; I can't see how anyone who enjoyed software development could like that.
Enjoying something and getting satisfaction out of it are two different things. I don't enjoy the act of coding. But I enjoy the feeling when I figure something out. I also think that having to solve novel puzzles as part of my job helps preserve my brain plasticity as I age. I'm not sure I'll get either of those from claude.
I feel zero sense of sadness about how things used to be. I feel like the change that sucked the most was when software engineering went from something that nerds did because they were passionate about programming, to techbros who were just in it for the money. We lost the idealism of the web a long time ago and the current swamp with apex reptiles like Zuckerberg is what we have now. It became all about the bottom line a long time ago.
The two emotions I personally feel are fear and excitement. Fear that the machines will soon replace me. Excitement about the things I can build now and the opportunities I’m racing towards. I can’t say it’s the most enjoyable experience. The combo is hellish on sleep. But the excitement balances things out a bit.
Maybe I’d feel a sense of sadness if I didn’t feel such urgency to try and ride this tsunami instead of being totally swept away by it.
It's not a pure dichotomy though. I have always been both, and slowly mixing in agentic coding for work has left me some new headspace to do "trad" programming on side projects at home.
I love the exciting ideation phase, I love putting together the puzzle that makes the product work, and I also take pride in the craft.
I think the argument is "a bit too nice," it isn't a binary, motivations are complicated and sometimes both feelings coexist.
If I reflect for a moment about why I personally got into tech, I can find at least a few different reasons:
- because I like solving problems. It's sad that the specific types of problems I used to do are gone, but it's exciting that there are new ones.
- because I like using my skills to help other people. It's sad that one specific way I could do that is now less effective, but it's exciting that I can use my knowledge in new ways.
- because I like doing something where I can personally make a difference. Again, it cuts both ways.
The divide was never invisible and there has always been at least three camps.
The "make-it-go" people couldn't make anything go back then either. They build ridiculous unmaintainable code with or without AI. Since they are cowboys that don't know what they're doing, they play the blame game and kiss a ton of ass.
The "craft-lovers" got in the way just as much with their endless yak shaving. They now embrace AI because they were just using "craft" as an excuse for why they didn't know what they were doing. They might be slightly more productive now only because they can argue with themselves instead of the rest of the team.
The more experienced and pragmatic people have always been forced to pick up the slack. If they have a say, they will keep scope narrow for the other two groups so they don't cause much damage. Their use of AI is largely limited to google searches like it always was.
Eh, it also feels like a classic "maybe we somehow have enough perspective on this watershed moment while it's happening to explain it with a simplistic dichotomy". Even this piece interrogates the feeling of "loss" and teases out multiple aspects to it, but settles on a tl;dr of "yep, dichotomy". There's more axes here too, where that feeling can depend on what you're building, who you're building it with, time and position in your career, etc etc.
(I'll admit, though, that this also smells to me a bit too much like introvert/extrovert, or INTP/INTJ/etc so maybe I'm being reflexively rejective)
I think the article misunderstands completely. "Craft" coders are chasing results too — we're just chasing results that last and that can be built upon. I've been in this game for a while, and a major goal of every single good programmer I've known has been to make themselves obsolete. Yes, I enjoyed meticulous hand crafted assembly, counting cycles and packing bits, but nobody had to talk me into using compilers. Yes, I've spent many fruitful hours writing basic CRUD apps but now that's easily done by libraries/frameworks I'm not eager to go back. Memory management, type systems, higher level languages, no-/low-code systems that completely remove me from some parts of the design loop, etc etc etc. All great: the point of computer programming is to have the computer do things so we don't have to.
I think the real divide we're seeing is between people who saw software as something that is, fundamentally, improvable and understandable; and people who saw it as a mysterious roadblock foisted upon them by others, that cannot really be reasoned about or changed. And oddly, many of the people in the second category use terminology from the first, but fundamentally do not believe that the first category really exists. (Fair enough; I was surprised at the second category.) It's not about intelligence or whatever, it's a mindset or perspective thing.
You can use gen AI entirely in the spirit of craft. For instance if you need to consume, implement or extend some open source software you can load it up in an agent IDE and ask “How do I?” questions or “how is it that?” questions that put you on a firm footing.
> I was afraid the puzzle-solving was over. But it wasn't—it just moved up a level.
The craft can move up a level too. You still can make decisions about the implementation, which algorithms to use, how to combine them, how and what to test -- essentially crafting the system at a higher level. In a similar sense, we lost the hand-crafting of assembly code as compilers took over, and now we're losing the crafting of classes and algorithms to some extent, but we still craft the system -- what and how it does its thing, and most importantly, why.
Do we even want a bunch of people contributing slop upstream when (assuming it does anything worthwhile in the first place) somebody has to actually review/correct/document that code?
A handful of well intentioned slop piles might be manageable, but AI enables spewing garbage at an unprecedented scale. When there's a limited amount of resources to expend on discussing, reviewing, fixing, and then finally accepting contributions a ton of AI generated contributions from random people could bring development to a halt.
I’ve heard this thesis a lot, but it’s almost always from the result chasers.
It doesn’t resonate with me because I am a result chaser. I like woodworking because I like building something that never existed before. I don’t mind using a CNC router or a 3 printer to help me out. I don’t care about the process, I care about the result. But I care deeply about the quality of the result.
I don’t care about the beauty of the code, but I do care that nearly every app I load takes longer than it did 15 years ago. I do care that my HomePod tells my wife it’s having trouble connecting to iPhone every 5th time she adds something to the grocery list. I care that my brokerage website is so broken that I actually had to call tech support who told me that they know it’s broken and you have to add a parameter to go back to the old version to get it to work.
I care that when I use the Claude desktop app it sometimes gives me a pop up with buttons that I can’t click on.
I’ve used Claude and Cursor enough to have what I think are valid opinions on AI assisted coding. Coding is not the bottleneck to produce a qualify product. Understanding the problem is the biggest bottleneck. Knowing what to build and what not to build. The next big one is convincing everyone around you of that (sometime this takes even more time). After that, it’s obsessively spending time iterating on something until it’s flawless. Sometimes that’s tweaking an easing value until the animation feels just right. Sometimes that’s obsessing over performance, and sometimes it’s freezing progress until you can make the existing app bulletproof.
AI doesn’t help me with these. At least not much. Mostly because the time I spend coding is time I spend understanding, diagnosing, and perfecting. Not the code. The product.
It does help crank out one off tools. It does help me work in unfamiliar code bases, or code bases where for whatever reason I care more about velocity than quality. It helps me with search. It helps me rubber duck.
All of those things does boost my productivity I think, but maybe somewhere in the order of 10% all in.
> AI doesn’t help me with these. At least not much. Mostly because the time I spend coding is time I spend understanding, diagnosing, and perfecting. Not the code. The product.
It can actually help a lot here too.
In fact I rarely have AI author or edit code, but I have it all time researching, finding edge cases I didn't think about, digging into dependencies code, finding ideas or alternative approaches. My usage is 90% of the time assisting with information gathering, criticizing (I have multiple reviewer skills with different personas, and I have multiple LLMs run them), refining, reviewing.
Even when it comes to product stuff, many of my clients have complicated business logic. Talking multi-tenant-company warehouse software where each process is drastically different and complexity balloons fast even for a single one of them. It helps to connect the dots between different sources of information (old Jira task, discord dumps, confluence, codebase, etc).
And it can iteratively test and find edge cases in applications too, same as you would do manually by taking control of the browser and testing the most uncommon paths.
I would do much less without this assistance.
I really don't get why people focus so much on the least empowering part (code), where it actually tends to balloon complexity quick or overwhelm you with so much content and edits that you can't have the energy to follow while maintaining quality.
Yeah I’ve used it for that and it use useful. But it’s kind of like listening to a math audiobook vs working out math problems. I still need to work out the math problems to really understand what’s going on.
I’m also nervous about the inevitable cognitive decline of relying on AI to explain everything to me.
You added several paragraphs after I responded, but I never said it was black and white. I said it wasn’t sufficient. By which I mean you still need to read the code, and that making changes to the code is even more beneficial for understanding.
> Coding is not the bottleneck to produce a qualify product.
I've been saying this too. The 10x engineer stuff simply cannot make sense unless previously you were spending 90%+ of your day just writing coding and now that's dropped to single digits because AI can generate it. If you spent 20-30% of your day coding before and the rest thinking about the problem, thinking about good UX, etc, then AI coding assistances mathematically cannot make you a 10x engineer. At a push they might make you a 2x engineer.
Given this I think I realised something earlier about my own output... I'm probably just a unusually good coder. I've been doing this since I was a kid so writing and reading code is basically second nature to me. When I was a young teen I would literally take my laptop on holiday with me just so I could write code - I was just that kind of person.
So I've basically always been the strongest or one of the strongest coders on any team I've been on. I very rarely have to think about how to do something in code. It's hard to think back to a time when code was a significant bottleneck for me.
However, my output was never really faster than anyone else when it come to shipping, but the quality of my output has always been wayyy higher. And I think that was because I always spent a lot more time thinking and iterating to get the best result, while other people I work with spent far more time writing code and just trying to get something they could PR.
My problem now is that the people I work, some of whom can't even read code, are able to spit out thousands of lines of code a day. So this forcing me to cut corners just to keep up with the rest of the team.
6-12 months ago I'd get at least 2-3 calls a day from people on my team asking for help to write some code. Now they just ask the AI. I haven't had someone ask me a coding related question in months at this point.
I find this frustrating to be honest. I'm seeing bad decisions everywhere in the code. For example, often a change is hard because it's a bad idea. Perhaps a page on a website doesn't really look great on mobile or desktop. Previously you would have had to think about how you could come up with a good responsive design and implement the right breakpoints. But now people can just ask Claude Code to build a completely different page for mobile, so they do. For a human that would be a huge effort, even if someone who stupid enough to think that was a good idea they probably be forced to do something thats easier to maintain and implement, but an AI? Who cares. It works. The AI isn't going to tell you no.
I know the quality of code is dropping. I see the random bugs from people clearly not understanding what Claude is writing, but if they can just ask the AI to fix it, does it even matter?
> All of those things does boost my productivity I think, but maybe somewhere in the order of 10% all in.
I'm very much like you. AI doesn't really boost my productivity at all but that's because I care about what I build and don't find coding hard. So AI doesn't really offer me anything. All it's doing is making people who don't care what their building and don't care about the quality of their code more productive. And putting me under pressure to trade quality for velocity.
You seem to be conflating code quality with product integrity.
All those problems are caused by business decisions, not the developers. You do make a good point though that AI may enable more people to build their own when they can.
The term code quality is overloaded and not really worth discussing without defining exactly what we mean.
But yes many of those problems were caused by business decisions. But engineers are perfectly cable of creating those problems on their own. If an engineer doesn’t realize that the function they called buffers messages in memory because someone made a wrapper function around sendAsync() and called it send(), that’s a code quality issue not a business issue (except as in the broader sense where every problem is ultimately a business issue).
Or if an engineer writes a naive implementation of some algorithm and adds a spinner so that an operation takes 5s to finish when it could be instantaneous if they’d thought about the problem more.
> These are real feelings about real losses. I'm not here to argue otherwise.
I'll argue it. Technically, there's no loss IMO, only gain. Craft lovers can still lovingly craft away, even if they have to do it on their own time instead of on their now-AI-dominated day job, just like in ye olde days. Nothing's stopping them.
But now result chasers can get results faster in their chasing. Or get results at all. I'm a proud result chaser now making serious progress on various projects that I've had on ice anywhere from months to years and occasionally lamented not having time/energy for them. And I also note my stable of tools, for both AI-related dev and other things, has grown greatly in a short period of time. And I'm loving it.
> Before AI, both camps were doing the same thing every day. Writing code by hand. Using the same editors, the same languages
Throughout college I would see a pretty stark divide, where most people would use vscode on mac or on Windows + WSL. But there was a small minority who would spend alot of time 'tinkering' (e.g, experiment with OS like nix/gentoo, or tweaking their dev environment). Maybe i'm misunderstanding what a 'craft lover' means here but it seemed to me, at the time, that the latter camp had more technical depth just based on conversation. Can't speak to the result in terms of test scores. Though it would be interesting to see any data on that if it exists.
For instance, the ones that look at it from an economics perspective, security perspective, long term maintainability perspective and so on. For each of these there are pros and cons.
all this so people like us can do a job that wasn't that hard to begin with and was actually very comfortable all things considered, just a tiny bit easier in a way that isn't even measurable.
Ah, management without managing. Its depressing and engaging at the same time. Depressing because palace intrigue is exhausting and fraught with peril. Engaging because I love explaining things to people and watching everything click into place for them (see the 1 of 10k xkcd comic).
I am enjoying crafting really good requirements documents. I use an iterative process. The implementation is the test of the requirements document. If it's not right, I adjust the doc, discard that implementation, and try again.
After years of working at startups I’ve long since abandoned any notion of craft at work. I have developed a very keen sense for harmfully cutting corners though, and unreviewed AI code (or unreasonably large PRs - defined by a size you can’t comfortably review) is absolutely cutting corners. It’s nothing to do with craft and everything to do with both correctness and incurring massive amounts of future debt.
Yep it's not 'result chasers' but people who want to get credit while avoiding real work. And when their stuff breaks they are always too busy with something else or moved on to another project.
This better matches my experiences and feelings than the divide which the author discusses. The craft is in the entire building, not specifically the coding aspects. I want to do a great job building a house, and if AI helps or even completes some aspects while meeting standards that's awesome. The problem is it hasn't yet shown it can be trusted as a sub trade, and we've got people outsourcing the entire project to an army of agents. The result looks a lot like the condo I bought a few years ago.
I'm a craft lover but I like using the Ai for tedious tasks. Just today it tracked down a library conflict in a pom that from experience would have taken a day of trial and error.
I don't know how I feel about this. I started programming in 1979.
I went for a job in AI in the late 1980s and realised from the bonkers spin of the company founders that it really wasn't the 5 to 10 years away as I was being told. I went looking something that was going to deliver a result.
I came back to it maybe 6 years ago when while on the bench at a consultancy. I got into trying to do various Kaggle challenges. Then the boss got the bug and wanted to predict the answers to weird spurious money-making questions. I tried but even when there was good data, I didn't know how to do better anyone else. When there wasn't good data it just produced complete shit.
Since then the world has changed. Everything I touch has AI built in. And it's really good. When you don't know your way around something or you've got stuck it really gets you moving again. Yeah, if it regurgitates a stupid negative example from the documentation as if it is "the way to do it", you just ignore it because you have already read that.
Now, every week I'm subjected to lectures by people who don't know how to code about how productive AI is going to make me. Working in the financial sector every Californian pipe dream seems to be an imperative, but all must verified by an adult. My IDE tries to insert all sorts of crap into my production code as I type, and then I'm supposed to be allow it to generate my unit tests.
I know it will get better, but will it be another 5 to 10 years?
>I think recognizing which kind of grief you're feeling is the actually useful thing here. If you're mourning the loss of the craft itself—the texture of writing code, the satisfaction of an elegant solution—that's real, and no amount of "just adapt" addresses it. You might need to find that satisfaction somewhere else, or accept that work is going to feel different. Frankly, we've been lucky there's been a livelihood in craft up to now.
The blog post is all about being clear-eyed about the source of grief, but doesn't seem to articulate that it's the livelihood that's gone, not the craft. There's never been a better time to practice the craft itself.
Well, yeah, that's what a lot of folks are sad about - they can't practice the craft concurrently with the livelihood quite as much. But if you don't have a livelihood, you probably don't have as much space for craft at all.
The important thing to remember is that for a large number of people (in the US), "work" is a place where they do things that they hate for eight hours a day, for people they hate (surveys routinely show between 40% and 60% of people are "satisfied" with their jobs). Those of us who are in the tech industry because we like actually programming computers (the "craft-lovers", in the parlance of this blog post) have been lucky enough to have jobs where where we get to actually do something we enjoy (even if it's intermingled with meetings and JIRA). If AI slop really is the future and programming becomes as rare of a job as hand-building wood furniture, then most of us are going to be living the normal experience of capitalism in a way that we are probably not well-prepared for.
Personally, I have noticed that I still produce substantially more and better code than the people at my company spending all day writing prompts, so I'm not too worried yet, but it seems plausible at some point that a machine that stole every piece of software ever written will be able to reliably turn a few hundred watt-hours of of electricity into a hallucination-free PR.
I agree some people go to work to work, and claude is find / good for them, but I feel that characterization of us who are loving claude is disingenuous. I’m a creative, while I loved coding and honed my craft, it was creating that always had me hooked. Claude is creating on steroids. Not to mention, it can help you massively improve your code cleanliness. All of the little nice-to-have features, the cleanups, the high unit test coverage, nagging bug fixes, etc., they’re all trivial to do now.
It’s not the same as writing code, but it’s fun.
If your coworkers can’t outpace your code output they’re either not using opus4.6 or they aren’t really trying.
It’s pretty easy to slam 20 PRs a day with some of them being complex changes. The hardest part is testing the diffs, but people are figuring that out too.
I have a suite of Claude skills all about craftsmanship. Refactoring, renaming, deconstructing god classes, detecting deleted code, etc. I've never written better, more readable, more maintainable code in my life than I have with Claude. All of that is trivial and hardly takes any time at all to accomplish.
Before moving to agentic AI, I thought I'd miss the craftsmanship aspect. Not at all. I get great satisfaction out of having AI write readable, maintainable code with me in the driver's seat.
> it can help you massively improve your code cleanliness. All of the little nice-to-have features, the cleanups, the high unit test coverage, nagging bug fixes, etc., they’re all trivial to do now.
It can help if you write poor code without it, probably
High unit test coverage only means something if you carefully check those tests, and if everything was tested
I think it's more granular than this, though. I also like to "make computer do thing" and have enjoyed using AI. But I also like building systems, optimizing systems. I find AI is a great partner in that. I can churn out prototypes more quickly, iterate on them more quickly etc. That also applies intra-system level. I might have a theory about how a different data structure or caching layer will affect application performance. It's now so much faster to test those kind of theories, and actually building good scaffolding around them to test them scientifically.
Yes, sometimes I can also ask AI to evaluate things at the system level and it often has surprisingly good insights, but that is usually a collaboration where our powers combined comes up with a better solution. I enjoy that process, too.
I do sympathize with the people "in mourning". I feel like this is really about how your identify is tied up in what you do. I have generally identified as a command line wizard. The xkcd of the guy flying in with "perl" very much speaks to me. But AI absolutely crushes at this. It's not that useful a skill anymore. Now I identify more as a local AI expert instead :D
> I feel like this is really about how your identify is tied up in what you do.
This is it for me. One thing that's important to my identify making things. And I have a lot of trouble saying I made a thing that I asked someone (or something) else to make for me.
I know you're going to say, "But I'm making things, too!" However...
I could crank out a project a day with Claude Code and slap them all up on GitHub for my green squares, and I could say that I made them all.
Just like I could crank out a novel a day with ChatGPT and say that I made them all.
Or I could use it to write 100 blog posts a day and say that I made them all.
In all those cases, I caused things to get made. But did I make them? I don't feel like I can honestly say I did. (And the copyright office is starting to have a thing or two to say about it, as well.)
This is what I struggle with. I like making things.
As a capitalist, sure, your cash is good with me. Tell me where to shoot and I'll shoot. But in terms of keeping my soul fed, it's a tough one.
Brilliant engineers, among the best software craftsmen out there are using AI daily and speeding up their processes.
The author of Redis, antirez, stated a month ago he spent 2 weeks on Redis tinkering with LLMs...and it was just design phase, not a single line of code was authored. The ability to interrogate LLMs and have them criticize his ideas and edge cases sped up his process by month.
He also used LLMs successfully to find multiple issues in Redis that would've took him longer to do without.
I myself spend with AI way more time tinkering and gathering information than authoring code.
Am I a craft lover or a result chaser?
But sure, let's keep everything in the divide conservative vs liberal, black and white, craftsman vs vibe coder...give me a break..
It's too bad it has so much grating LLM-voice. I don't think you typed all the LLM-isms, and they make it hard to know how much to trust that the rest is what you intended to convey.
I'm a 23+ year dev; among the highest level ICs in my org.
It's still craft, its just that the craft is different. I don't write *.ts, *.cs files anymore; I write *.md files that other devs are using, that we're using as guardrails, that ensures that we minimize the slop while increasing speed and basically lift every developers level up by several notches.
I went from building one kind of framework/platform level artifact to another type of framework/platform level artifact.
If one's perspective is that it's just a shift in what "craft" means, then it's still craft. I'm still building systems; just a different kind of system.
You're using it as a 'super compiler', effectively a code generator and your .md file is the new abstraction level at which you code.
But there is a price to pay: the code that you generate is not the code that you understand and when things go pear shaped you will find that that deterministic element that made compilers so successful is missing from code generated from specs dumped into an AI. If you one-shot it you will find that the next time you do this your code may come out quite different if it isn't a model that you maintain. It may contain new bugs or revive old ones. It may eliminate chunks of the code and you'll never know and so on.
There is a reason that generated code always had a bit of a smell to it and AI generated code is no different. How much time do you spend on verifying that it actually does what's written on the tin?
Do you write your own tests? Do you let the AI write the tests and the code? Are you familiar with the degree to which AIs can be manipulated to do stuff that you thought they weren't supposed to? (A friend of mine just proved this to his boss by bribing an AI with a 'nice batch of pure random data' to put a piece of unreviewed code into production by giving itself the privileges required to do so...)
Quality and consistency are going up, not down. Partially because the agents follow the guidance much more closely than humans do and there is far less variance. Shortcuts that a human would make ("I'll just write a one-off here"), the agent does not...so long as our rules guide it properly ("Let me find existing patterns in the codebase.").
Part of it is the investment in docs we've made. Part of it is that we were already meticulous about commenting code. It turns out that when the agents stumble on this code randomly, it can read the comments (we can tell because it also updates them in PRs when it makes changes).
We are also delivering the bulk of our team level capabilities via remote MCP over HTTP so we have centralized telemetry via OTEL on tool activation, docs being read by the agents, phantom docs the agent tries to find (we then go and fill in those docs).
There are some studies about maintaining attention over longer periods of time when there is no action required. It will be difficult to keep that up forever so beware of review fatigue and bake in some measures to ensure that attention does not diminish over time.
The point of reviews is that the process of reviews is a feedback cycle where we can identify where our docs are short. We then immediately update the docs to reflect the correction.
Over enough time, this gap closes and the need for reviews goes down. This is what I've noticed as we've continued to improve the docs: PRs have stabilized. Mid-level devs that just months ago were producing highly variant levels of quality are now coalescing on a much higher, much more consistent level of output.
There were a lot of pieces that went into this. We created a local code review skill that encodes the exact heuristics the senior reviewers would use and we ask the agent to run this in AGENTS.md. We have an MCP server over HTTP that we use to deliver the docs so we can monitor centralized telemetry.
The objective is that at some point, there will be enough docs and improved models that the need for human reviews decreases while quality of code reaches a steady state that is more consistent than any human team of varying skill level could produce.
One thing we've done is to decouple the docs from the codebase to make it easier to update the docs and immediately distribute updates orthogonal to the lifecycle of a PR.
(I'll have a post at some point that goes into some of what we are doing and the methodology.)
> Partially because the agents follow the guidance much more closely than humans do and there is far less variance.
Ouch. Managing human coders has been described as herding cats (with some justice). Getting humans to follow standards is... challenging. And exhausting.
Getting AIs to do so... if you get the rules right, and if the tool doesn't ignore the rules, then you should be good. And if you're not, you still have human reviews. And the AI doesn't get offended if you reject the PR because it didn't follow the rules.
This is actually one of the best arguments for AIs that I have seen.
Yes, as I mentioned in my other replies, what I've seen is that quality has gone up and coalesced around a much higher bar with far less variance than before as we've refined our docs and tooling.
In some cases, it was "instant"; dev's MCP server connected to our docs was down -> terrible PR. We fix the MCP connection and redo the PR -> instantly follows the guides we have in place for best practices.
>A friend of mine just proved this to his boss by bribing an AI with a 'nice batch of pure random data' to put a piece of unreviewed code into production by giving itself the privileges required to do so...
Okay that's pretty hilarious. Everyone has a vice!
Came here to say something similar. For me, the craft aspect is now even more exciting because I can craft more ambitious things without getting bogged down in the details. For me, refining my conceptual model, drawing diagrams, finding the right way to think about something was the craft.
Maybe that's another way of saying: I was trained as a designer, and now the distinction between design (read: architecture, service-design, product, ux, cx) and programming is blurring.
Heck yeah! Love that way of putting it. Agree. Now there’s more time to focus on making the right architecture and carrying it out. It’s no longer a days long task to do a big refactor to remove code smells.
Every little minor dispute can be split into some arbitrary dichotomy which is vaguely defensible. Not interesting.
Twelve years ago I would have the bright idea of why not make a little, just a tiny little (what I would call now) preprocessor for Java which does the same thing in less characters and is clearer. Everyone would love it. Of course no one loved it. Well, I never implemented it. Because I got some sense: you can’t just make tiny little preprocessors, a little code generation here and there, just code-generate this and tweak after the fact. Right? It’s not principled.
You can cook up a dichotomy. Good for you. I think the approach is just space age technology meets Stone Age mindset. It’s Flintstone Engineering. It’s barely even serious.
I am not offended that you took my craft. I am offended that you smear paint on the wall with three hundred parallel walls and painters and pick the best one. Or whatever Rube Setup is the thing that will take over the world as of thirty minutes ago.
Make something rock solid like formal verification with LLM assist (or LLM with formal verification assist?). Something that a “human” can understand (at this point maybe only the CEO is left). Something that is understandable, deterministic.
I might be out of a job. But I will not be offended. And I will respect it.
I'm a bit in the middle. I enjoy the craft but I also seek and enjoy the result.
The thing about AI is that you don't have to use it for everything. Like any other tool you can use it as much as you'd like. Even though I like the craft, I find myself really enjoying the use of AI to do things like boilerplate code and simple tests. I hate crafting verbose grunt work, so I have AI do that. This in turn leaves me more time to do the interesting work.
I also enjoy using AI to audit, look for bugs, brainstorm, and iterate on ideas. When an idea is solid and fleshed out I'll craft the hard and interesting parts and AI-generate the boring parts.
People who say directing an AI is just "moving up another level of abstraction" are missing the point that it's a completely different kind of work. Everything from machine code to Haskell is a predictable deductive logical system, whereas AIs are not.
Lots of mentions of the term mourning... As they say in my country, don't sell the skin until you kill the bear.
All I'm seeing around me is people dropping best practices in a FOMO driven push for speed: let's stop reviews, let's drive 5 agents in parallel, let's not even look at the code!
This is going to blow up.
Only after we pick up the remains we'll find a more sustainable approach for AI usage. I suspect that version will still require crafters.
If we end up in a place where the craft truly is dead, then congratulations, your value probably just dropped to zero. Everyone who's been around startup culture knows the running jokes about those 'I have a great idea, I just need someone to code it' guys. Now you're one, and you'll find how much ideas are worth.
> If we end up in a place where the craft truly is dead, then congratulations, your value probably just dropped to zero
I think the craft is going to die and am not thrilled about it. I dont feel like there is a contradiction there
> If we end up in a place where the craft truly is dead, then congratulations, your value probably just dropped to zero
I think, then that the value of all knowledge work will have dropped to zero. Software engineering is, to my mind, “intelligence complete.” If you can do it with knowledge work, you can have software do it.
That’s not the point of nor the reason for knowledge work.
The fundamental mechanism of knowledge work is people. They haven’t changed at all. And what they need to understand and learn hasn’t changed. All the agents in the world and all of the methane guzzling data centres can’t tell you what to write in the specification nor if what the computer has generated faithfully implements that specification.
Yep. Most knowledge work is about coordination between people and transporting the right information to the people that are thinking and the people that are doing (not always separate groups, but can be one group switching mode). You need people because there's a lot of shared context between individuals in society that is not encoded anywhere.
A year ago everyone was sure these things couldn’t write functional code. A few months ago people started saying they need to be operated by people who could otherwise write the code.
Like… the guy in Office Space who took specifications from the customer via a secretary and gave those to the engineers (and we know what happened to him).
But I’m not sure that’s a thing, at least for long, either. The original super power of these things wasn’t that they could write code. It was that they could very competently extract meaning from natural language, debug what you were saying from the terrible way you expressed it, and still formulate competent answers.
That doesn’t sound like a comfortable place for former devs to sit for the next few years.
Exactly right.
I have an idea for an “evidence editor.” Claude is waiting for me to tell it exactly what I want this thing to be. But I don’t know. I haven’t figured out how to square the various circles, even in my fantasies. Until I do, Claude sits and waits. And waits…
The beginnings of that sustainable approach is already out there: https://boristane.com/blog/how-i-use-claude-code/
These bear related sayings always make me laugh. The one I was told by a Russian: “don’t argue over how to skin the bear before you’ve killed it”
A couple of guys at work have been raving about Claude. How quick they get stuff done, how great the code is, how working any other way is a waste of time.
I just had the misfortune today to wade into one of their codebases. It's 60k lines of code for something that should have been simple, and it's an absolute fucking mess. I'm gonna have to rip out most of it and start over just to get it to do what we actually need it to do.
I use LLMs, they come in handy, and I use agents, but this "have agents do everything" nonsense is a disaster, and it's only going to get worse.
On the upside I'm getting paid to fix this shit show.
> On the upside I'm getting paid to fix this shit show.
A lot of my career has been this, not due to choice but circumstance. Startups write terrible code, in general. Enterprises write terrible code. I’ve worked with both. If it becomes important enough, someone has to fix it at some point.
Current AI models seem to be job security machines for that kind of work.
>This is going to blow up.
We are way past wringing our hands over agentic engineering. Every startup and all fast moving companies are onboard. They don't hand code anymore. There will not be some code quality crisis that will stop everyone in their tracks. I'm trying to cope with this too, but I don't think the best path is praying for failure.
Just out of the popularity of the claim, I’ll bite.
Both big tech and startups are now full of people working at 10x, features are written as fast as PMs can think them, monoliths self heal with agents buzzing over them.
10x means 10 times the outcomes in a given amount of time, so did you see the last iOS version pack a decade worth of features in a single release?
Do you remember when meta moved their backend to rust in a month?
What about Microsoft software not having a single bug in a year?
Yeah, me neither.
I didn't say anything about increased productivity or 10x. Feel free to revise your strawman.
Fair, let’s revise it then.
If not productivity, what’s the result AI is getting that is disruptive enough to make our previous work obsolete?
At 11 this morning, I wanted to both debug an issue and take a meeting before lunch. Before AI, I would have had to just start debugging after lunch, there wouldn't have been enough time to do both. But now I had Claude debug the issue concurrently with the meeting. Its answer didn't actually make sense (I still do think I'm smarter than Claude, although the gap is narrowing!), but it showed enough of its work that I could make a good guess about what was really going on, and when I asked it to check my hypothesis I got back from lunch with some debug logs that confirmed I'd found the bug.
I can easily believe that, I agree Claude has applications.
I am disputing the idea that this is enough of a game changer to make us mourn our now lost craft. Also, I’m mentioning that we’ve discovered a world of footguns dressed as shortcuts, which we’re not taking proper care of.
First, your experience was required for that story to have a happy ending. Second, we both know someone else could probably have gone with Claude’s senseless hypothesis, asked for a fix and sent it for review. This last part is becoming pretty universal.
If there’s no increased productivity then what’s the point in spending all the money?
I didn't say there was or there wasn't. They just don't get to infer that I did and then attack that as my position.
10x is definitely possible at a startup level. I suppose not in a big tech world (seems obvious to me, and it's not like development speed was the bottleneck there either, right?)
You can choose not to believe what I say (and I genuinely understand if you do), and I can simply keep on doing it. I'm not taking it out of thin air either. tuesday I did in 8h the work I scheduled for roughly 65h. Ok, so maybe it's not 10x, maybe it's 8x, same ball park.
And that's only talking about development. If I now get into other aspects....I have just spent the last 1.5 hours creating an incredibly detailed backcatalog of tasks and epics. This is the most detailed I have ever done so in my life and it has been working very well. It's like we merged the good of highly-detailed waterfall with the speed of agile.
Tee-hee: Watergile. (I'm sure some expert in the field will let me know I have coined a new term for something that very much has a name; excuse my ignorance in advance).
Nonetheless, I did this all by talking to the computer which is interfacing with my project management tools, the project documentation, and the project code. Full context on everything. In the past, I would have taken 3 or 4 days to create the same amount of tasks with a vaguely similar amount of detail. But, in truth, I wouldn't have spent so much time putting this love into the craft (!!) of planning a project, because it would exhaust me and feel like a waste of time.
Don't get me wrong, I totally see shit code being thrown everywhere by inferior AI models or people who can't tame the beasts, but the right people in my life are _clearly_ building out more, better tested code, and actually built with more care. Maybe it's not at the line-by-line level, but it certainly is from the end-product result (thinking of the actual end-user). I accept your mileage may vary -- this is my very personal experience.
Maybe it'll stop happening, who knows. Maybe price will be prohibitive, or maybe we'll have such an avalanche of ideas that weren't worth building that everyone will be overwhelmed and take a step back. Or maybe we won't develop juniors into the seniors of tomorrow. Or maybe everything will indeed implode once products are large enough that the original development speed can't be maintained anymore and expectations are mismatched.
What I do know is that it is definitely happening in my world, and I haven't had this much fun since I was a little kid learning to code.
Do you have testable hypothesis for how the 10x will manifest? I.e., is there a way we could (coarsely) measure this in a year or two from now?
That's a very good question. Some metrics I'm watching or considering watching right now:
- Amount of leads we're taking in (per unit of time)
- Sprint velocity changes (task complexity should stay roughly the same with AI, and team velocity increase — we've been seeing this happening)
- Hire rates (more sales people, less developers?)
- Number of projects per unit of time (of similar dimension, hard to measure)
- Length of "bugfixing buffer" before big releases (we've actually been noticing this go down)
- Another way of saying it is: number of bugs, or bugs per feature
- Drift between planned execution time and actual execution time (we've been delivering early...but I guess we'll soon adjust our estimates...or maybe not, who knows?)
- Spend on AI models
- I can't measure this, but I can sort of "feel it": but the overall feedback we get from clients, the feeling we get from them.
- Number of tests (tests have skyrocketed. Can't be sure about the quality, but, hey, it's a metric)
- Feature turnaround time (how long since "feature is proposed" until it's actually implemented)
- documentation to code ratio (not sure what we'll make of it, but there's a somewhat worrying trend here)
- team balance: is everyone slowly becoming fullstack? Do we feel that those who aren't are significantly affecting development speed? if so, that indicates that the other ones are somehow moving faster
I can't really think of any others, but I'm sure they exist.
The future will not be evenly distributed. You can't expect to see it in the productivity of the industry as a whole, or even the productivity of a large company. You might be able to see it in a medium-sized team if you measure carefully.
I think I agree with you. I didn't mean to imply that big corps will build faster -- in fact I said I don't expect them to right now, and I'm not so sure that will change.
What I believe is that early prototype development and pivoting is insanely fast now. And if you find excellent engineers who are also great product people, and then pair them with people who have truly great ideas, many wonderful new products will emerge.
So, it sounds like measuring directly in individual individuals or companies might be tough. (Unless the company is medium-sized?) Maybe we could look for broader trends in the economy and beyond. What sorts of companies will this manifest in? I.e. mostly "tech" companies, or beyond?
> I don't think the best path is praying for failure. Embrace it
"Embrace failure" is exactly the attitude every company is going to take. They've already been working at bottoming out our expectations.
We should have been running companies out of business with regulations and abandonment when their human-written software leaked our private data to criminals, or when their untested forced-updates shut down our systems and sent our IT teams scrambling, or when their unoptimized code forced us to upgrade our hardware or negated any performance gains we should have seen from investing in upgrades.
The quality, reliability, and security of the software we all use and depend on is going to nosedive, and companies already know they can get away with it. They aren't going to start caring about how we feel about that now. "Pay more and settle for less" is where we are today. "Embrace Failure" is the future we're sprinting towards.
That's absolutely not true. The places that have embraced "agentic engineering" are mostly garbage factories, and lots of places, including plenty of startups and fast-moving companies are staying off of this trend. I recognize that most of the people on this site are just trying to self-promote for their own gig, but the level of misinformation is sometimes just staggering.
Want something to be terrified of?
I work at a massive health care company. They're 100% on the AI bandwagon and are putting AI everywhere they can. Billing, Software, DevOps, everywhere. If you think you can give an Agent some information and have go to work for some user, its 100% on the table for the company to do and either a) then outsource the rest offshore or b) lay the person off or shrink the department to increase the bottom line.
Your healthcare, right now, is being offloaded to AI agents and bots and this is only the beginning.
I literally just sat through the annual “choose your healthcare” plan bullshit and the “meeting” was literally one of the Hr people pulling up a power point narrated by “AI”. You could tell in the first ten seconds.
You’d think our plans would be cheaper given they’re offloading all this work to agents they don’t have to pay a salary to…right?
>lots of places, including plenty of startups and fast-moving companies are staying off of this trend.
Provide some examples then? Everyone who is all in on agentic code are pretty vocal about it. Who is declaring the opposite stance? Anyone?
Both claims are hyperbole.
Reality remains in the middle, but there are plenty of examples of either extreme right now.
It is not just startups or small companies embracing agentic engineering… Stripe published blog posts about their autonomous coding agents. Amazon is blowing up production because they gave their agents access to prod. Google and Microsoft develop their own agentic engineering tools. It’s not just tech companies either, massive companies are frequently announcing their partnerships with OpenAI or Anthropic.
You can’t just pretend it’s startups doing all the agentic engineering. They’re just the ones pushing the boundaries on best practices the most aggressively.
Outwards communication and inside results tend to differ vastly. I’ve heard some true horror stories already from companies who claim they’re doing amazing things with great results. You should be especially on guard if it’s a publicly traded company, selling AI usage is necessary to appease the market (and thereby C-level stock value).
>Outwards communication and inside results tend to differ vastly.
This is a good call out, but I'm talking to a lot of friends at other companies. So my perspective is informed by both news and personal anecdote.
Sure, it goes both ways, I’m having great results at the startup I’m working at too.
Well nobody has had to pay the tech debt yet on the last 6 months of that insanity. I think the age-old SWE best practices will still hold in time.
I was still skeptical at the start of this year, but there seems to be a shift underway. Found the StrongDM Dark Factory docs in Feb and they've netted novel results that have been inspiring enough to keep studying and practicing.
https://factory.strongdm.ai/techniques
https://factory.strongdm.ai/products/attractor
If you've found better or ancillary resources, please share.
Wow! My productivity increased 100x after i stated listening to ad bots.
Apologies if my non-LLM optimized messaging didn't resonate with you. Have you found anything else which produces decent results consistently?
This sounds right to me:
> Before AI, both camps were doing the same thing every day. Writing code by hand. Using the same editors, the same languages, the same pull request workflows. The craft-lovers and the make-it-go people sat next to each other, shipped the same products, looked indistinguishable. The motivation behind the work was invisible because the process was identical.
Helps explain why some people are delighted to have AI write code for them while others are unhappy that the part they enjoyed so much has been greatly reduced.
Similar note from Kellan (a clear member of the make-it-go group) in https://laughingmeme.org/2026/02/09/code-has-always-been-the... :
> That feeling of loss though can be hard to understand emotionally for people my age who entered tech because we were addicted to feeling of agency it gave us. The web was objectively awful as a technology, and genuinely amazing, and nobody got into it because programming in Perl was somehow aesthetically delightful.
I think the real divide is over quality and standards.
We all have different thresholds for what is acceptable, and our roles as engineers typically reflect that preference. I can grind on a single piece of code for hours, iterating over and over until I like the way it works, the parameter names, etc.
Other people do not see the value in that whatsoever, and something that works is good enough. We both are valuable in different ways.
Also, theres the pace of advancement of the models. Many people formed their opinions last year, and the landscape has changed a lot. There’s also some effort requires in honing your skill using them. The “default” output is average quality, but with some coaxing higher quality output is easily attained.
I’m happy people are skeptical though, there are a lot of things that do require deep thought, connecting ideas in new ways, etc., and LLMs aren’t good at that in my experience.
> I think the real divide is over quality and standards.
I think there are multiple dimensions that people fall on regarding the issue and it's leading to a divide based on where everyone falls on those dimensions.
Quality and standards are probably in there but I think risk-tolerance/aversion could be behind some how you look at quality and standards. If you're high on risk-taking, you might be more likely to forego verifying all LLM-generated code, whereas if you're very risk-averse, you're going to want to go over every line of code to make sure it works just right for fear of anything blowing up.
Desire for control is probably related, too. If you desire more control in how something is achieved, you probably aren't going to like a machine doing a lot of the thinking for you.
I think this is a false dichotomy because which approach is acceptable depends heavily on context, and good engineers recognize this and are capable of adapting.
Sometimes you need something to be extremely robust and fool-proof, and iterating for hours/days/weeks and even months might make sense. Things that are related to security or money are good examples.
Other times, it's much more preferable to put something in front of users that works so that they start getting value from it quickly and provide feedback that can inform the iterative improvements.
And sometimes you don't need to iterate at all. Good enough is good enough. Ship it and forget about it.
I don't buy that AI users favor any particular approach. You can use AI to ship fast, or you can use it to test, critique, refactor and optimize your code to hell and back until it meets the required quality and standards.
> nobody got into it because programming in Perl was somehow aesthetically delightful.
To this day I remember being delighted by Perl on a regular basis. I wasn't concerned with the aesthetics of it, though I was aware it was considered inscrutable and that I could read & write it filled me with pride. So yea, programming Perl was delightful.
I strongly disagree. There's always been two camps ... on everything!
Emacs vs. vi. Command-line editor vs. IDE. IntelliJ vs. VS Code. I could do like twenty more of these: dev teams have always split on technology choices.
But, all of those were rational separations. Emacs and vi, IntelliJ and VS Code ... they're all viable options, so they boil down to subjective preference. By definition, anything subjective will vary between different humans.
What makes AI different to me is the fear. Nobody decided not to use emacs because they were afraid it was going to take their job ... but a huge portion of the anti-AI crowd is motivated by irrational fear, related to that concern.
what about the fear is irrational?
It doesn’t help that the CEOs of these companies are hyping up the fear. It’s no wonder people are afraid when the people making the products are spouting prophecies of doom.
A huge portion of the pro-AI crowd is motivated by irrational hype and delusion.
LLMs are not a tool like an editor or an IDE, they make up code in an unpredictable way; I can't see how anyone who enjoyed software development could like that.
Enjoying something and getting satisfaction out of it are two different things. I don't enjoy the act of coding. But I enjoy the feeling when I figure something out. I also think that having to solve novel puzzles as part of my job helps preserve my brain plasticity as I age. I'm not sure I'll get either of those from claude.
I feel zero sense of sadness about how things used to be. I feel like the change that sucked the most was when software engineering went from something that nerds did because they were passionate about programming, to techbros who were just in it for the money. We lost the idealism of the web a long time ago and the current swamp with apex reptiles like Zuckerberg is what we have now. It became all about the bottom line a long time ago.
The two emotions I personally feel are fear and excitement. Fear that the machines will soon replace me. Excitement about the things I can build now and the opportunities I’m racing towards. I can’t say it’s the most enjoyable experience. The combo is hellish on sleep. But the excitement balances things out a bit.
Maybe I’d feel a sense of sadness if I didn’t feel such urgency to try and ride this tsunami instead of being totally swept away by it.
It's not a pure dichotomy though. I have always been both, and slowly mixing in agentic coding for work has left me some new headspace to do "trad" programming on side projects at home.
I love the exciting ideation phase, I love putting together the puzzle that makes the product work, and I also take pride in the craft.
I think the argument is "a bit too nice," it isn't a binary, motivations are complicated and sometimes both feelings coexist.
If I reflect for a moment about why I personally got into tech, I can find at least a few different reasons:
- because I like solving problems. It's sad that the specific types of problems I used to do are gone, but it's exciting that there are new ones.
- because I like using my skills to help other people. It's sad that one specific way I could do that is now less effective, but it's exciting that I can use my knowledge in new ways.
- because I like doing something where I can personally make a difference. Again, it cuts both ways.
I'm sure most people would cite similar examples.
The divide was never invisible and there has always been at least three camps.
The "make-it-go" people couldn't make anything go back then either. They build ridiculous unmaintainable code with or without AI. Since they are cowboys that don't know what they're doing, they play the blame game and kiss a ton of ass.
The "craft-lovers" got in the way just as much with their endless yak shaving. They now embrace AI because they were just using "craft" as an excuse for why they didn't know what they were doing. They might be slightly more productive now only because they can argue with themselves instead of the rest of the team.
The more experienced and pragmatic people have always been forced to pick up the slack. If they have a say, they will keep scope narrow for the other two groups so they don't cause much damage. Their use of AI is largely limited to google searches like it always was.
Eh, it also feels like a classic "maybe we somehow have enough perspective on this watershed moment while it's happening to explain it with a simplistic dichotomy". Even this piece interrogates the feeling of "loss" and teases out multiple aspects to it, but settles on a tl;dr of "yep, dichotomy". There's more axes here too, where that feeling can depend on what you're building, who you're building it with, time and position in your career, etc etc.
(I'll admit, though, that this also smells to me a bit too much like introvert/extrovert, or INTP/INTJ/etc so maybe I'm being reflexively rejective)
I think the article misunderstands completely. "Craft" coders are chasing results too — we're just chasing results that last and that can be built upon. I've been in this game for a while, and a major goal of every single good programmer I've known has been to make themselves obsolete. Yes, I enjoyed meticulous hand crafted assembly, counting cycles and packing bits, but nobody had to talk me into using compilers. Yes, I've spent many fruitful hours writing basic CRUD apps but now that's easily done by libraries/frameworks I'm not eager to go back. Memory management, type systems, higher level languages, no-/low-code systems that completely remove me from some parts of the design loop, etc etc etc. All great: the point of computer programming is to have the computer do things so we don't have to.
I think the real divide we're seeing is between people who saw software as something that is, fundamentally, improvable and understandable; and people who saw it as a mysterious roadblock foisted upon them by others, that cannot really be reasoned about or changed. And oddly, many of the people in the second category use terminology from the first, but fundamentally do not believe that the first category really exists. (Fair enough; I was surprised at the second category.) It's not about intelligence or whatever, it's a mindset or perspective thing.
You can use gen AI entirely in the spirit of craft. For instance if you need to consume, implement or extend some open source software you can load it up in an agent IDE and ask “How do I?” questions or “how is it that?” questions that put you on a firm footing.
> I was afraid the puzzle-solving was over. But it wasn't—it just moved up a level.
The craft can move up a level too. You still can make decisions about the implementation, which algorithms to use, how to combine them, how and what to test -- essentially crafting the system at a higher level. In a similar sense, we lost the hand-crafting of assembly code as compilers took over, and now we're losing the crafting of classes and algorithms to some extent, but we still craft the system -- what and how it does its thing, and most importantly, why.
And contribute your changes back upstream, right?
Do we even want a bunch of people contributing slop upstream when (assuming it does anything worthwhile in the first place) somebody has to actually review/correct/document that code?
A handful of well intentioned slop piles might be manageable, but AI enables spewing garbage at an unprecedented scale. When there's a limited amount of resources to expend on discussing, reviewing, fixing, and then finally accepting contributions a ton of AI generated contributions from random people could bring development to a halt.
I’ve heard this thesis a lot, but it’s almost always from the result chasers.
It doesn’t resonate with me because I am a result chaser. I like woodworking because I like building something that never existed before. I don’t mind using a CNC router or a 3 printer to help me out. I don’t care about the process, I care about the result. But I care deeply about the quality of the result.
I don’t care about the beauty of the code, but I do care that nearly every app I load takes longer than it did 15 years ago. I do care that my HomePod tells my wife it’s having trouble connecting to iPhone every 5th time she adds something to the grocery list. I care that my brokerage website is so broken that I actually had to call tech support who told me that they know it’s broken and you have to add a parameter to go back to the old version to get it to work.
I care that when I use the Claude desktop app it sometimes gives me a pop up with buttons that I can’t click on.
I’ve used Claude and Cursor enough to have what I think are valid opinions on AI assisted coding. Coding is not the bottleneck to produce a qualify product. Understanding the problem is the biggest bottleneck. Knowing what to build and what not to build. The next big one is convincing everyone around you of that (sometime this takes even more time). After that, it’s obsessively spending time iterating on something until it’s flawless. Sometimes that’s tweaking an easing value until the animation feels just right. Sometimes that’s obsessing over performance, and sometimes it’s freezing progress until you can make the existing app bulletproof.
AI doesn’t help me with these. At least not much. Mostly because the time I spend coding is time I spend understanding, diagnosing, and perfecting. Not the code. The product.
It does help crank out one off tools. It does help me work in unfamiliar code bases, or code bases where for whatever reason I care more about velocity than quality. It helps me with search. It helps me rubber duck.
All of those things does boost my productivity I think, but maybe somewhere in the order of 10% all in.
> AI doesn’t help me with these. At least not much. Mostly because the time I spend coding is time I spend understanding, diagnosing, and perfecting. Not the code. The product.
It can actually help a lot here too.
In fact I rarely have AI author or edit code, but I have it all time researching, finding edge cases I didn't think about, digging into dependencies code, finding ideas or alternative approaches. My usage is 90% of the time assisting with information gathering, criticizing (I have multiple reviewer skills with different personas, and I have multiple LLMs run them), refining, reviewing.
Even when it comes to product stuff, many of my clients have complicated business logic. Talking multi-tenant-company warehouse software where each process is drastically different and complexity balloons fast even for a single one of them. It helps to connect the dots between different sources of information (old Jira task, discord dumps, confluence, codebase, etc).
And it can iteratively test and find edge cases in applications too, same as you would do manually by taking control of the browser and testing the most uncommon paths.
I would do much less without this assistance.
I really don't get why people focus so much on the least empowering part (code), where it actually tends to balloon complexity quick or overwhelm you with so much content and edits that you can't have the energy to follow while maintaining quality.
Yeah I’ve used it for that and it use useful. But it’s kind of like listening to a math audiobook vs working out math problems. I still need to work out the math problems to really understand what’s going on.
I’m also nervous about the inevitable cognitive decline of relying on AI to explain everything to me.
> But it’s kind of like listening to a math audiobook vs working out math problems.
I don't see it that way.
I'm solving the math problem, but after coming with a solution I start asking for alternative approaches or formulas I don't even know about.
In fact, calculus is full of suck gotchas tricks or formulas, think of integrals or limits. Took who found those decades/centuries to find them.
It's not a black/white divide.
You added several paragraphs after I responded, but I never said it was black and white. I said it wasn’t sufficient. By which I mean you still need to read the code, and that making changes to the code is even more beneficial for understanding.
> Coding is not the bottleneck to produce a qualify product.
I've been saying this too. The 10x engineer stuff simply cannot make sense unless previously you were spending 90%+ of your day just writing coding and now that's dropped to single digits because AI can generate it. If you spent 20-30% of your day coding before and the rest thinking about the problem, thinking about good UX, etc, then AI coding assistances mathematically cannot make you a 10x engineer. At a push they might make you a 2x engineer.
Given this I think I realised something earlier about my own output... I'm probably just a unusually good coder. I've been doing this since I was a kid so writing and reading code is basically second nature to me. When I was a young teen I would literally take my laptop on holiday with me just so I could write code - I was just that kind of person.
So I've basically always been the strongest or one of the strongest coders on any team I've been on. I very rarely have to think about how to do something in code. It's hard to think back to a time when code was a significant bottleneck for me.
However, my output was never really faster than anyone else when it come to shipping, but the quality of my output has always been wayyy higher. And I think that was because I always spent a lot more time thinking and iterating to get the best result, while other people I work with spent far more time writing code and just trying to get something they could PR.
My problem now is that the people I work, some of whom can't even read code, are able to spit out thousands of lines of code a day. So this forcing me to cut corners just to keep up with the rest of the team.
6-12 months ago I'd get at least 2-3 calls a day from people on my team asking for help to write some code. Now they just ask the AI. I haven't had someone ask me a coding related question in months at this point.
I find this frustrating to be honest. I'm seeing bad decisions everywhere in the code. For example, often a change is hard because it's a bad idea. Perhaps a page on a website doesn't really look great on mobile or desktop. Previously you would have had to think about how you could come up with a good responsive design and implement the right breakpoints. But now people can just ask Claude Code to build a completely different page for mobile, so they do. For a human that would be a huge effort, even if someone who stupid enough to think that was a good idea they probably be forced to do something thats easier to maintain and implement, but an AI? Who cares. It works. The AI isn't going to tell you no.
I know the quality of code is dropping. I see the random bugs from people clearly not understanding what Claude is writing, but if they can just ask the AI to fix it, does it even matter?
> All of those things does boost my productivity I think, but maybe somewhere in the order of 10% all in.
I'm very much like you. AI doesn't really boost my productivity at all but that's because I care about what I build and don't find coding hard. So AI doesn't really offer me anything. All it's doing is making people who don't care what their building and don't care about the quality of their code more productive. And putting me under pressure to trade quality for velocity.
You seem to be conflating code quality with product integrity.
All those problems are caused by business decisions, not the developers. You do make a good point though that AI may enable more people to build their own when they can.
The term code quality is overloaded and not really worth discussing without defining exactly what we mean.
But yes many of those problems were caused by business decisions. But engineers are perfectly cable of creating those problems on their own. If an engineer doesn’t realize that the function they called buffers messages in memory because someone made a wrapper function around sendAsync() and called it send(), that’s a code quality issue not a business issue (except as in the broader sense where every problem is ultimately a business issue).
Or if an engineer writes a naive implementation of some algorithm and adds a spinner so that an operation takes 5s to finish when it could be instantaneous if they’d thought about the problem more.
> These are real feelings about real losses. I'm not here to argue otherwise.
I'll argue it. Technically, there's no loss IMO, only gain. Craft lovers can still lovingly craft away, even if they have to do it on their own time instead of on their now-AI-dominated day job, just like in ye olde days. Nothing's stopping them.
But now result chasers can get results faster in their chasing. Or get results at all. I'm a proud result chaser now making serious progress on various projects that I've had on ice anywhere from months to years and occasionally lamented not having time/energy for them. And I also note my stable of tools, for both AI-related dev and other things, has grown greatly in a short period of time. And I'm loving it.
> Before AI, both camps were doing the same thing every day. Writing code by hand. Using the same editors, the same languages
Throughout college I would see a pretty stark divide, where most people would use vscode on mac or on Windows + WSL. But there was a small minority who would spend alot of time 'tinkering' (e.g, experiment with OS like nix/gentoo, or tweaking their dev environment). Maybe i'm misunderstanding what a 'craft lover' means here but it seemed to me, at the time, that the latter camp had more technical depth just based on conversation. Can't speak to the result in terms of test scores. Though it would be interesting to see any data on that if it exists.
There are far more divides than just that one.
For instance, the ones that look at it from an economics perspective, security perspective, long term maintainability perspective and so on. For each of these there are pros and cons.
all this so people like us can do a job that wasn't that hard to begin with and was actually very comfortable all things considered, just a tiny bit easier in a way that isn't even measurable.
> a job that wasn't that hard to begin with
The more experience I get the harder the job seems tbh
Have you gotten to the part where you barely even get to write code anymore and just manage people's expectations full time yet?
Ah, management without managing. Its depressing and engaging at the same time. Depressing because palace intrigue is exhausting and fraught with peril. Engaging because I love explaining things to people and watching everything click into place for them (see the 1 of 10k xkcd comic).
I am enjoying crafting really good requirements documents. I use an iterative process. The implementation is the test of the requirements document. If it's not right, I adjust the doc, discard that implementation, and try again.
> I am enjoying crafting really good requirements documents.
Can you please come train the product people at my job? Maybe some of your love of the game will rub off on them.
After years of working at startups I’ve long since abandoned any notion of craft at work. I have developed a very keen sense for harmfully cutting corners though, and unreviewed AI code (or unreasonably large PRs - defined by a size you can’t comfortably review) is absolutely cutting corners. It’s nothing to do with craft and everything to do with both correctness and incurring massive amounts of future debt.
Yep it's not 'result chasers' but people who want to get credit while avoiding real work. And when their stuff breaks they are always too busy with something else or moved on to another project.
This better matches my experiences and feelings than the divide which the author discusses. The craft is in the entire building, not specifically the coding aspects. I want to do a great job building a house, and if AI helps or even completes some aspects while meeting standards that's awesome. The problem is it hasn't yet shown it can be trusted as a sub trade, and we've got people outsourcing the entire project to an army of agents. The result looks a lot like the condo I bought a few years ago.
> Before AI, both camps were doing the same thing every day. Writing code by hand. Using the same editors, the same languages,
Hell no. I, a craftsman, was going out of my way to use things like Haskell. I was very aware of the divide the entire time. The present is a relief.
I'm a craft lover but I like using the Ai for tedious tasks. Just today it tracked down a library conflict in a pom that from experience would have taken a day of trial and error.
I don't know how I feel about this. I started programming in 1979.
I went for a job in AI in the late 1980s and realised from the bonkers spin of the company founders that it really wasn't the 5 to 10 years away as I was being told. I went looking something that was going to deliver a result.
I came back to it maybe 6 years ago when while on the bench at a consultancy. I got into trying to do various Kaggle challenges. Then the boss got the bug and wanted to predict the answers to weird spurious money-making questions. I tried but even when there was good data, I didn't know how to do better anyone else. When there wasn't good data it just produced complete shit.
Since then the world has changed. Everything I touch has AI built in. And it's really good. When you don't know your way around something or you've got stuck it really gets you moving again. Yeah, if it regurgitates a stupid negative example from the documentation as if it is "the way to do it", you just ignore it because you have already read that.
Now, every week I'm subjected to lectures by people who don't know how to code about how productive AI is going to make me. Working in the financial sector every Californian pipe dream seems to be an imperative, but all must verified by an adult. My IDE tries to insert all sorts of crap into my production code as I type, and then I'm supposed to be allow it to generate my unit tests.
I know it will get better, but will it be another 5 to 10 years?
Are we 80% of the way there yet?
> Since then the world has changed. Everything I touch has AI built in. And it's really good.
Clearly you don’t use Amazon’s Alexa.
>I think recognizing which kind of grief you're feeling is the actually useful thing here. If you're mourning the loss of the craft itself—the texture of writing code, the satisfaction of an elegant solution—that's real, and no amount of "just adapt" addresses it. You might need to find that satisfaction somewhere else, or accept that work is going to feel different. Frankly, we've been lucky there's been a livelihood in craft up to now.
The blog post is all about being clear-eyed about the source of grief, but doesn't seem to articulate that it's the livelihood that's gone, not the craft. There's never been a better time to practice the craft itself.
Well, yeah, that's what a lot of folks are sad about - they can't practice the craft concurrently with the livelihood quite as much. But if you don't have a livelihood, you probably don't have as much space for craft at all.
It’s sad not because of AI itself but because of the companies behind AI: we are now paying for every single line of code we produce. That sucks
Pointy haired bosses be looking for results.
Engineers be loving the craft.
It's a dance, but AI is unfortunately looking at us like we're dancing, and meanwhile it's built a factory.
This reminds me of the divide between Role-players and Number-chasers in the once-upon-a-time MUD players communities.
The important thing to remember is that for a large number of people (in the US), "work" is a place where they do things that they hate for eight hours a day, for people they hate (surveys routinely show between 40% and 60% of people are "satisfied" with their jobs). Those of us who are in the tech industry because we like actually programming computers (the "craft-lovers", in the parlance of this blog post) have been lucky enough to have jobs where where we get to actually do something we enjoy (even if it's intermingled with meetings and JIRA). If AI slop really is the future and programming becomes as rare of a job as hand-building wood furniture, then most of us are going to be living the normal experience of capitalism in a way that we are probably not well-prepared for.
Personally, I have noticed that I still produce substantially more and better code than the people at my company spending all day writing prompts, so I'm not too worried yet, but it seems plausible at some point that a machine that stole every piece of software ever written will be able to reliably turn a few hundred watt-hours of of electricity into a hallucination-free PR.
I agree some people go to work to work, and claude is find / good for them, but I feel that characterization of us who are loving claude is disingenuous. I’m a creative, while I loved coding and honed my craft, it was creating that always had me hooked. Claude is creating on steroids. Not to mention, it can help you massively improve your code cleanliness. All of the little nice-to-have features, the cleanups, the high unit test coverage, nagging bug fixes, etc., they’re all trivial to do now.
It’s not the same as writing code, but it’s fun.
If your coworkers can’t outpace your code output they’re either not using opus4.6 or they aren’t really trying.
It’s pretty easy to slam 20 PRs a day with some of them being complex changes. The hardest part is testing the diffs, but people are figuring that out too.
I have a suite of Claude skills all about craftsmanship. Refactoring, renaming, deconstructing god classes, detecting deleted code, etc. I've never written better, more readable, more maintainable code in my life than I have with Claude. All of that is trivial and hardly takes any time at all to accomplish.
Before moving to agentic AI, I thought I'd miss the craftsmanship aspect. Not at all. I get great satisfaction out of having AI write readable, maintainable code with me in the driver's seat.
> it can help you massively improve your code cleanliness. All of the little nice-to-have features, the cleanups, the high unit test coverage, nagging bug fixes, etc., they’re all trivial to do now.
It can help if you write poor code without it, probably
High unit test coverage only means something if you carefully check those tests, and if everything was tested
> a machine that stole every piece of software.
Please, give it a rest. Nothing was stolen. Words have meanings.
Stealing has been used for copyright infringement since forever, it is the correct word
I think it's more granular than this, though. I also like to "make computer do thing" and have enjoyed using AI. But I also like building systems, optimizing systems. I find AI is a great partner in that. I can churn out prototypes more quickly, iterate on them more quickly etc. That also applies intra-system level. I might have a theory about how a different data structure or caching layer will affect application performance. It's now so much faster to test those kind of theories, and actually building good scaffolding around them to test them scientifically.
Yes, sometimes I can also ask AI to evaluate things at the system level and it often has surprisingly good insights, but that is usually a collaboration where our powers combined comes up with a better solution. I enjoy that process, too.
I do sympathize with the people "in mourning". I feel like this is really about how your identify is tied up in what you do. I have generally identified as a command line wizard. The xkcd of the guy flying in with "perl" very much speaks to me. But AI absolutely crushes at this. It's not that useful a skill anymore. Now I identify more as a local AI expert instead :D
> I feel like this is really about how your identify is tied up in what you do.
This is it for me. One thing that's important to my identify making things. And I have a lot of trouble saying I made a thing that I asked someone (or something) else to make for me.
I know you're going to say, "But I'm making things, too!" However...
I could crank out a project a day with Claude Code and slap them all up on GitHub for my green squares, and I could say that I made them all.
Just like I could crank out a novel a day with ChatGPT and say that I made them all.
Or I could use it to write 100 blog posts a day and say that I made them all.
In all those cases, I caused things to get made. But did I make them? I don't feel like I can honestly say I did. (And the copyright office is starting to have a thing or two to say about it, as well.)
This is what I struggle with. I like making things.
As a capitalist, sure, your cash is good with me. Tell me where to shoot and I'll shoot. But in terms of keeping my soul fed, it's a tough one.
I just do not want to deal with other people's AI-generated code.
Your AI can rewrite it to your own standards for free.
You can tell your AI to read their code, and create a new requirements document for a clean-room implementation.
Then you have your AI implement using your own requirements document.
That doesn't sound like it would help while reviewing a PR full of verbose slop in your day job.
In a review process?
Yow, submitter sure isn't shy with their bias. Maybe defang the title?
There's no divide.
Brilliant engineers, among the best software craftsmen out there are using AI daily and speeding up their processes.
The author of Redis, antirez, stated a month ago he spent 2 weeks on Redis tinkering with LLMs...and it was just design phase, not a single line of code was authored. The ability to interrogate LLMs and have them criticize his ideas and edge cases sped up his process by month.
He also used LLMs successfully to find multiple issues in Redis that would've took him longer to do without.
I myself spend with AI way more time tinkering and gathering information than authoring code.
Am I a craft lover or a result chaser?
But sure, let's keep everything in the divide conservative vs liberal, black and white, craftsman vs vibe coder...give me a break..
Author doesn't care about their blog writing as craft, either (it's been fed through an LLM.)
Sure, I ran the post past an LLM for some ideas on clarity and tightening it up - but I wrote, edited, and published it myself.
It's too bad it has so much grating LLM-voice. I don't think you typed all the LLM-isms, and they make it hard to know how much to trust that the rest is what you intended to convey.
I guess I have a grating LLM-voice, then, because I don't think it sounds particularly different than how I've written other posts.
Am I the only one who comes to the comments first to see if the blog/article is even worth reading?
The divide is a matter of perspective.
I'm a 23+ year dev; among the highest level ICs in my org.
It's still craft, its just that the craft is different. I don't write *.ts, *.cs files anymore; I write *.md files that other devs are using, that we're using as guardrails, that ensures that we minimize the slop while increasing speed and basically lift every developers level up by several notches.
I went from building one kind of framework/platform level artifact to another type of framework/platform level artifact.
If one's perspective is that it's just a shift in what "craft" means, then it's still craft. I'm still building systems; just a different kind of system.
You're using it as a 'super compiler', effectively a code generator and your .md file is the new abstraction level at which you code.
But there is a price to pay: the code that you generate is not the code that you understand and when things go pear shaped you will find that that deterministic element that made compilers so successful is missing from code generated from specs dumped into an AI. If you one-shot it you will find that the next time you do this your code may come out quite different if it isn't a model that you maintain. It may contain new bugs or revive old ones. It may eliminate chunks of the code and you'll never know and so on.
There is a reason that generated code always had a bit of a smell to it and AI generated code is no different. How much time do you spend on verifying that it actually does what's written on the tin?
Do you write your own tests? Do you let the AI write the tests and the code? Are you familiar with the degree to which AIs can be manipulated to do stuff that you thought they weren't supposed to? (A friend of mine just proved this to his boss by bribing an AI with a 'nice batch of pure random data' to put a piece of unreviewed code into production by giving itself the privileges required to do so...)
We have human reviews on every PR.
Quality and consistency are going up, not down. Partially because the agents follow the guidance much more closely than humans do and there is far less variance. Shortcuts that a human would make ("I'll just write a one-off here"), the agent does not...so long as our rules guide it properly ("Let me find existing patterns in the codebase.").
Part of it is the investment in docs we've made. Part of it is that we were already meticulous about commenting code. It turns out that when the agents stumble on this code randomly, it can read the comments (we can tell because it also updates them in PRs when it makes changes).
We are also delivering the bulk of our team level capabilities via remote MCP over HTTP so we have centralized telemetry via OTEL on tool activation, docs being read by the agents, phantom docs the agent tries to find (we then go and fill in those docs).
> We have human reviews on every PR.
There are some studies about maintaining attention over longer periods of time when there is no action required. It will be difficult to keep that up forever so beware of review fatigue and bake in some measures to ensure that attention does not diminish over time.
The point of reviews is that the process of reviews is a feedback cycle where we can identify where our docs are short. We then immediately update the docs to reflect the correction.
Over enough time, this gap closes and the need for reviews goes down. This is what I've noticed as we've continued to improve the docs: PRs have stabilized. Mid-level devs that just months ago were producing highly variant levels of quality are now coalescing on a much higher, much more consistent level of output.
There were a lot of pieces that went into this. We created a local code review skill that encodes the exact heuristics the senior reviewers would use and we ask the agent to run this in AGENTS.md. We have an MCP server over HTTP that we use to deliver the docs so we can monitor centralized telemetry.
The objective is that at some point, there will be enough docs and improved models that the need for human reviews decreases while quality of code reaches a steady state that is more consistent than any human team of varying skill level could produce.
One thing we've done is to decouple the docs from the codebase to make it easier to update the docs and immediately distribute updates orthogonal to the lifecycle of a PR.
(I'll have a post at some point that goes into some of what we are doing and the methodology.)
> Partially because the agents follow the guidance much more closely than humans do and there is far less variance.
Ouch. Managing human coders has been described as herding cats (with some justice). Getting humans to follow standards is... challenging. And exhausting.
Getting AIs to do so... if you get the rules right, and if the tool doesn't ignore the rules, then you should be good. And if you're not, you still have human reviews. And the AI doesn't get offended if you reject the PR because it didn't follow the rules.
This is actually one of the best arguments for AIs that I have seen.
Yes, as I mentioned in my other replies, what I've seen is that quality has gone up and coalesced around a much higher bar with far less variance than before as we've refined our docs and tooling.
In some cases, it was "instant"; dev's MCP server connected to our docs was down -> terrible PR. We fix the MCP connection and redo the PR -> instantly follows the guides we have in place for best practices.
>A friend of mine just proved this to his boss by bribing an AI with a 'nice batch of pure random data' to put a piece of unreviewed code into production by giving itself the privileges required to do so...
Okay that's pretty hilarious. Everyone has a vice!
There is a chapter two to the story but I don't want to out my friend. You never know who reads HN.
Came here to say something similar. For me, the craft aspect is now even more exciting because I can craft more ambitious things without getting bogged down in the details. For me, refining my conceptual model, drawing diagrams, finding the right way to think about something was the craft.
Maybe that's another way of saying: I was trained as a designer, and now the distinction between design (read: architecture, service-design, product, ux, cx) and programming is blurring.
Heck yeah! Love that way of putting it. Agree. Now there’s more time to focus on making the right architecture and carrying it out. It’s no longer a days long task to do a big refactor to remove code smells.
Every little minor dispute can be split into some arbitrary dichotomy which is vaguely defensible. Not interesting.
Twelve years ago I would have the bright idea of why not make a little, just a tiny little (what I would call now) preprocessor for Java which does the same thing in less characters and is clearer. Everyone would love it. Of course no one loved it. Well, I never implemented it. Because I got some sense: you can’t just make tiny little preprocessors, a little code generation here and there, just code-generate this and tweak after the fact. Right? It’s not principled.
You can cook up a dichotomy. Good for you. I think the approach is just space age technology meets Stone Age mindset. It’s Flintstone Engineering. It’s barely even serious.
I am not offended that you took my craft. I am offended that you smear paint on the wall with three hundred parallel walls and painters and pick the best one. Or whatever Rube Setup is the thing that will take over the world as of thirty minutes ago.
Make something rock solid like formal verification with LLM assist (or LLM with formal verification assist?). Something that a “human” can understand (at this point maybe only the CEO is left). Something that is understandable, deterministic.
I might be out of a job. But I will not be offended. And I will respect it.
I'm a bit in the middle. I enjoy the craft but I also seek and enjoy the result.
The thing about AI is that you don't have to use it for everything. Like any other tool you can use it as much as you'd like. Even though I like the craft, I find myself really enjoying the use of AI to do things like boilerplate code and simple tests. I hate crafting verbose grunt work, so I have AI do that. This in turn leaves me more time to do the interesting work.
I also enjoy using AI to audit, look for bugs, brainstorm, and iterate on ideas. When an idea is solid and fleshed out I'll craft the hard and interesting parts and AI-generate the boring parts.
People who say directing an AI is just "moving up another level of abstraction" are missing the point that it's a completely different kind of work. Everything from machine code to Haskell is a predictable deductive logical system, whereas AIs are not.
Strong agree. Needs another pass or two at editing though, some painful LLM-os sticking out there :'(
Which bits? And don't say the em-dashes because I've been over-using them since high school