Rolling the ladder up behind us

Rolling the ladder up behind us Published on 2025-06-20 , 5674 words, 21 minutes to read Who will take over for us if we don't train the next generation to replace us? A critique of craft, AI, and the legacy of human expertise. A picture of two patches of wild grass bifurcated by a retaining pond. - Photo by Xe Iaso, Canon EOS R6 Mark 2, unknown lens Cloth is one of the most important goods a society can produce. Clothing is instrumental for culture, expression, and for protecting one's modesty. Historically, cloth was one of the most expensive items on the market. People bought one or two outfits at most and then wore them repeatedly for the rest of their lives. Clothing was treasured and passed down between generations the same way we pass jewelry down between generations. This cloth was made in factories by highly skilled weavers. These weavers had done the equivalent of PhD studies in weaving cloth and used state of the art hardware to do it. As factories started to emerge, they were able to make cloth so much more cheaply than skilled weavers ever could thanks to inventions like the power loom. Power looms didn't require skilled workers operating them. You could even staff them with war orphans, which there was an abundance of thanks to all the wars. The quality of the cloth was absolutely terrible in comparison, but there was so much more of it made so much more quickly. This allowed the price of cloth to plummet, meaning that the wages that the artisans made fell from six shillings a day to six shillings per week over a period of time where the price of food doubled. Mind you, the weavers didn't just reject technological progress for the sake of rejecting it. They tried to work with the ownership class and their power looms in order to produce the same cloth faster and cheaper than they had before. For a time, it did work out, but the powers that be didn't want that. They wanted more money at any cost. At some point, someone had enough and decided to do something about it. Taking up the name Ned, he led a movement that resulted in riots, destroying factory equipment, and some got so bad they had to call the army in to break them up. Townspeople local to those factory towns were in full support of Ned's followers. Heck, even the soldiers sent to stop the riots ended up seeing the points behind what Ned's followers were doing and joined in themselves. The ownership class destroyed the livelihood of the skilled workers so that they could make untold sums of money producing terrible cloth that people would turn their one-time purchase of clothing into a de-facto subscription that they had to renew every time their clothing wore out. Now we have fast fashion and don't expect our clothing to last more than a few years. I have a hoodie from AWS Re:Invent in 2022 that I'm going to have to throw out and replace because the sleeves are dying. We only remember them as riots because their actions affected those in power. This movement was known as the Luddites, or the followers of Ned Ludd. The word "luddite" has since shifted meaning over time and is now understood as "someone who is against technological development". The Luddites were not against technology like the propaganda from the ownership class would have you expect, they fought against how it was implemented and the consequences of its rollout. They were skeptical that the shitty cloth that the power loom produced would be a net benefit to society because it meant that customers would inevitably have to buy their clothes over and over again, turning a one-time purchase into a subscription. Would that really benefit consumers or would that really benefit the owners of the factories? Nowadays the Heritage Crafts Association of the United Kingdom lists many forms of weaving as Endangered or Critically Endangered crafts, meaning that those skills are either at critical risk of dying out without any "fresh blood" learning how to do it, or the last generation of artisans that know how to do that craft are no longer teaching new apprentices. All that remains of that expertise is now contained in the R&D departments of the companies that produce the next generations of power looms, and whatever heritage crafts practitioners remain. Remember the Apollo program that let us travel to the moon? It was mostly powered by the Rocketdyne F1 engine. We have all of the technical specifications to build that rocket engine. We know all the parts you need, all the machining you have to do, and roughly understand how it would be done, but we can't build another Rocketdyne F1 because all of the finesse that had been built up around manufacturing it no longer exists. Society has moved on and we don't have expertise in the tools that they used to make it happen. What are we losing in the process? We won't know until it's gone. We're going to run out of people with the word "Senior" in their title As I've worked through my career in computering, I've noticed a paradox that's made me uneasy and I haven't really been able to figure out why it keeps showing up: the industry only ever seems to want to hire people with the word Senior in their title. They almost never want to create people with the word Senior in their title. This is kinda concerning for me. People get old and no longer want to or are able to work. People get sick and become disabled. Accidental deaths happen and remove people from the workforce. A meme based on the format where the dog wants to fetch the ball but doesn't want to give the ball to the human to throw it, but with the text saying 'Senior?', 'Train Junior?', and 'No train junior, only hire senior'. If the industry at large isn't actively creating more people with the word Senior in their title, we are eventually going to run out of them. This is something that I want to address with Techaro at some point, but I'm not sure how to do that yet. I'll figure it out eventually. The non-conspiratorial angle for why this is happening is that money isn't free anymore and R&D salaries are no longer taxable business expenses in the US, so software jobs that don't "produce significant value" are more risky to the company. So of course they'd steal from the future to save today. Sounds familiar, doesn't it? Cadey Is this how we end up losing the craft of making high quality code the same way we lost the craft of weaving high quality cloth? However there's another big trend in the industry that concerns me: companies releasing products that replace expertise with generative AI agents that just inscrutably do the thing for you. This started out innocently enough - it was just better ways to fill in the blanks in your code. But this has ballooned and developed from better autocomplete to the point where you can just assign issues to GitHub Copilot and have the issue magically get solved for you in a pull request. Ask the AI model for an essay and get a passable result in 15 minutes. At some level, this is really cool. Like, think about it. This reduces toil and drudgery to waiting for half an hour at most. In a better world I would really enjoy having a tool like this to help deal with the toil work that I need to do but don't really have the energy to. Do you know how many more of these essays would get finished if I could offload some of the drudgery of my writing process to a machine? We are not in such a better world. We are in a world where I get transphobic hate sent to the Techaro sales email. We are in a world where people like me are intentionally not making a lot of noise so that we can slide under the radar and avoid attention by those that would seek to destroy us. We are in a world where these AI tools are being pitched as the next Industrial Revolution, one where foisting our expertise away into language models is somehow being framed as a good thing for society. There's just one small problem: who is going to be paid and reap the benefits from this change as expectations from the ownership class change? A lot of the ownership class only really experiences the work product outputs of what we do with computers. They don't know the struggles involved with designing things such as the user getting an email on their birthday. They don't want to get pushback on things being difficult or to hear that people want to improve the quality of the code. They want their sparkle emoji buttons to magically make the line go up and they want them yesterday. We deserve products that aren't cheaply made mass produced slop that incidentally does what people want instead of high quality products that are crafted to be exactly what people need, even if they don't know they need it. Additionally, if this is such a transformational technology, why are key figures promoting it by talking down to people? Why wouldn't they be using this to lift people up? Aoi Isn't that marketing? Fear sells a lot better than hope ever will. Amygdala responses are pretty strong right? So aren't a lot of your fears of the technology really feeding into the hype and promoting the technology by accident? Cadey I don't fear the power loom. I fear the profit expectations of the factory owners. Vibe coding is payday loans for technical debt As a technical educator, one of the things that I want to imprint onto people is that programming is a skill you can gain and that you too can both program things and learn how to program things. I want there to be more programmers out there. What I am about to say is not an attempt to gatekeep the skill and craft of computering; however, the ways that proponents of vibe coding are going about it are simply not the way forward to a sustainable future. About a year ago, Cognition teased an AI product named Devin, a completely automated software engineer. You'd assign Devin tasks in Slack or Jira and then it would spin up a VM and plod its way through fixing whatever you asked it to. This demo deeply terrified me, as it was nearly identical to a story I wrote for the Techaro lore: Protos. The original source of that satire was experience working at a larger company that shall remain unnamed where the product team seemed to operate under the assumption that the development team had a secret "just implement that feature button" and that we as developers were working to go out of our way to NOT push it. Devin was that "implement that feature" button the same way Protos mythically did. From what I've seen with companies that actually use Devin, it's nowhere near actually being useful and usually needs a lot of hand-holding to do anything remotely complicated, thank God. The thing that really makes me worried is that the ownership class' expectations about the process of developing software are changing. People are being put on PIPs for not wanting to install Copilot. Deadlines come faster because "the AI can write the code for you, right?" Twitter and Reddit contain myriads of stories of "idea guys" using Cursor or Windscribe to generate their dream app's backend and then making posts like "some users claim they can see other people's stuff, what kind of developer do I need to hire for this?" Follow-up posts include gems such as "lol why do coders charge so much???" By saving money in the short term by producing shitty software that doesn't last, are we actually spending more money over time re-buying nearly identical software after it evaporates from light use? This is the kind of thing that makes Canada not allow us to self-identify as Engineers, and I can't agree with their point more. Vibe Coding is just fancy UX Vibe coding is a distraction. It's a meme. It will come. It will go. Everyone will abandon the vibe coding tools eventually. My guess is that a lot of the startups propping up their vibe coding tools are trying to get people into monthly subscriptions as soon as possible so that they can mine passive income as their more casual users slowly give up on coding and just forget about the subscription. I'm not gonna lie though, the UX of vibe coding tools is top-notch. From a design standpoint it's aiming for that subtle brilliance where it seems to read your mind and then fill in the blanks you didn't even know you needed filled in. This is a huge part of how you can avoid the terror of the empty canvas. If you know what you are doing, an empty canvas represents infinite possibilities. There's nothing there to limit you from being able to do it. You have total power to shape everything. In my opinion, this is a really effective tool to help you get past that fear of having no ground to stand on. This helps you get past executive dysfunction and just ship things already. That part is a good thing. I genuinely want people to create more things with technology that are focused on the problems that they have. This is the core of how you learn to do new things. You solve small problems that can be applied to bigger circumstances. You gradually increase the scope of the problem as you solve individual parts of it. I want more people to be able to do software development. I think that it's a travesty that we don't have basic computer literacy classes in every stage of education so that people know how the machines that control their lives work and how to use them to their advantage. Sure it's not as dopaminergic as TikTok or other social media apps, but there's a unique sense of victory that you get when things just work. Sometimes that feeling you get when things Just Work™ is the main thing that keeps me going. Especially in anno dominium two thousand and twenty five. The main thing I'm afraid of is people becoming addicted to the vibe coding tools and letting their innate programming skills atrophy. I don't know how to suggest people combat this. I've been combating it by removing all of the automatic AI assistance from my editor (IE: I'll use a language server, but I won't have my editor do fill-in-the-middle autocomplete for me), but this isn't something that works for everyone. I've found myself more productive without it there and asking a model for the missing square peg to round hole when I inevitably need some toil code made. I ended up not shipping that due to other requirements, but you get what I'm going at. The "S" in MCP stands for Security The biggest arguments I have against vibe coding and all of the tools behind it boil down to one major point: these tools have a security foundation of sand. Most of the time when you install and configure a Model Context Protocol (MCP) server, you add some information to a JSON file that your editor uses to know what tools it can dispatch with all of your configuration and API tokens. These MCP servers run as normal OS processes with absolutely no limit to what they can do. They can easily delete all files on your system, install malware into your autostart, or exfiltrate all your secrets without any oversight. Oh, by the way, that whole "it's all in one JSON file with all your secrets" problem? That's now seen as a load-bearing feature so that scripts can automatically install MCP servers for you. You don't even need to get expertise in how the tools work! There's a MCP server installer MCP server so that you can say "Hey torment nexus, install GitHub integration for me please" and then it'll just do it with no human oversight or review on what you're actually installing. Seems safe to me! What could possibly go wrong? If this is seriously the future of our industry, I wish that the people involved would take one trillionth of an iota of care about the security of the implementation. This is the poster child for something like the WebAssembly Component Model. This would let you define your MCP servers with strongly typed interfaces to the outside world that can be granted or denied permissions by users with strong capabilities. Combined with the concept of server resources, this could let you expand functionality however you wanted. Running in WebAssembly means that the no MCP server can just read ~/.ssh/id_ed25519 and exfiltrate your SSH key. Running in WebAssembly means that it can't just connect to probably-not-malware.lol and then evaluate JavaScript code with user-level permissions on the fly. We shouldn't have to be telling developers "oh just run it all in Docker". We should have designed this to be fundamentally secure from the get-go. Personally, I only run MCP ecosystem things when contractually required to. Even then, I run it in a virtual machine that I've already marked as known compromised and use separate credentials not tied to me. Do with this information as you will. I had a lot of respect for Anthropic before they released this feculent bile that is the Model Context Protocol spec and initial implementations to the public. It just feels so half-baked and barely functional. Sure I don't think they expected it to become the Next Big Meme™, but I thought they were trying to do things ethically above board. Everything I had seen from Anthropic before had such a high level of craft and quality, and this was such a huge standout. We shouldn't have to be placing fundamental concerns like secret management or sandboxing as hand-waves to be done opt-in by the user. They're not gonna do it, and we're going to have more incidents where Cursor goes rogue and nukes your home folder until someone cares enough about the craft of the industry to do it the right way. Everyone suffers so the few can gain I have a unique view into a lot of the impact that AI companies have had across society. I'm the CEO of Techaro, a small one-person startup that develops Anubis, a Web AI Firewall Utility that helps mitigate the load of automated mass scraping so that open source infrastructure can stay online. I've had sales calls with libraries and universities that are just being swamped by the load. There's stories of GitLab servers eating up 64 cores of high-wattage server hardware due to all of the repeated scraping over and over in a loop. I swear a lot of this scraping has to be some kind of dataset arbitrage or something, that's the only thing that makes sense at this point. And then in the news the AI companies claim "oh no we're just poor little victorian era orphans, we can't possibly afford to fairly compensate the people that made the things that make our generative AI models as great as they are". When the US copyright office tried to make AI training not a fair use, the head of that office suddenly found themselves jobless. Why must these companies be allowed to take everything without recourse or payment to the people that created the works that fundamentally power the models? The actual answer to this is going to sound a bit out there, but stay with me: they believe that we're on the verge of creating artificial superintelligence; something that will be such a benevolent force of good that any strife in the short term will ultimately be cancelled out by the good that is created as a result. These people unironically believe that a machine god will arise and we'd be able to delegate all of our human problems to it and we'll all be fine forever. All under the thumb of the people that bought the GPUs with dollars to run that machine god. As someone that grew up in a repressed environment full of evangelical christianity, I recognize this story instantly: it's the second coming of Christ wrapped in technology. Whenever I ask the true believers entirely sensible questions like "but if you can buy GPUs with dollars, doesn't that mean that whoever controls the artificial superintelligence thus controls everyone, even if the AI is fundamentally benevolent?" The responses I get are illuminating. They sound like the kinds of responses that evangelicals give when you question their faith. Artists suffer first Honestly though, the biggest impact I've seen across my friends has been what's happened to art commissions. I'm using these as an indicator for how the programming industry is going to trend. Software development is an art in the same vein as visual/creative arts, but a lot of the craft and process that goes into visual art is harder to notice because it gets presented as a flat single-dimensional medium. Sometimes it can take days to get something right for a drawing. But most of the time people just see the results of the work, not the process that goes into it. This makes things like prompting "draw my Final Fantasy 14 character in Breath of the Wild" with images as references and getting a result in seconds look more impressive. If you commissioned a human to get a painting like this: An AI-generated illustration of my Final Fantasy 14 character composited into a screenshot of Breath of the Wild. Generated by GPT-4o through the ChatGPT interface. Inputs were a screenshot of Breath of the Wild and reference photos of my character. It'd probably take at least a week or two as the artist worked through their commission queue and sent you in-progress works before they got the final results. By my estimates between the artists I prefer commissioning, this would cost somewhere between 150 USD and 500 EUR at minimum. Probably more when you account for delays in the artistic process and making sure the artist is properly paid for their time. It'd be a masterpiece that I'd probably get printed and framed, but it would take a nonzero amount of time. If you only really enjoy the products of work and don't understand/respect any of the craftsmanship that goes into making it happen, you'd probably be okay with that instantly generated result. Sure the sun position in that image doesn't make sense, the fingers have weird definition, her tail is the wrong shape, it pokes out of the dress in a nonsensical way (to be fair, the reference photos have that too), the dress has nonsensical shading, and the layering of the armor isn't like the reference pictures, but you got the result in a minute! A friend of mine runs an image board for furry art. He thought that people would use generative AI tools as a part of their workflows to make better works of art faster. He was wrong, it just led to people flooding the site with the results of "wolf girl with absolutely massive milkers showing her feet paws" from their favourite image generation tool in every fur color imaginable, then with different characters, then with different anatomical features. There was no artistic direction or study there. Just an endless flood of slop that was passable at best. Sure, you can make high quality art with generative AI. There's several comic series where things are incredibly temporally consistent because the artist trained their own models and took the time to genuinely gain expertise with the tools. They filter out the hallucination marks. They take the time to use it as a tool to accelerate their work instead of replacing their work. The boards they post it to go out of their way to excise the endless flood of slop and by controlling how the tools work they actually get a better result than they got by hand, much like how the skilled weavers were able to produce high quality cloth faster and cheaper with the power looms. We are at the point where the artists want to go and destroy the generative image power looms. Sadly, they can't even though they desperately want to. These looms are locked in datacentres that are biometrically authenticated. All human interaction is done by a small set of trusted staff or done remotely by true believers. I'm afraid of this kind of thing happening to the programming industry. A lot of what I'm seeing with vibe coding leading to short term gains at the cost of long term toil is lining up with this. Sure you get a decent result now, but long-term you have to go back and revise the work. This is a great deal if you are producing the software though; because that means you have turned one-time purchases into repeat customers as the shitty software you sold them inevitably breaks, forcing the customer to purchase fixes. The one-time purchase inevitably becomes a subscription. We deserve more in our lives than good enough. Stop it with the sparkle emoji buttons Look, CEOs, I'm one of you so I get it. We've seen the data teams suck up billions for decades and this is the only time that they can look like they're making a huge return on the investment. Cut it out with shoving the sparkle emoji buttons in my face. If the AI-aided product flows are so good then the fact that they are using generative artificial intelligence should be irrelevant. You should be able to replace generative artificial intelligence with another technology and then the product will still be as great as it was before. When I pick up my phone and try to contact someone I care about, I want to know that I am communicating with them and not a simulacrum of them. I can't have that same feeling anymore due to the fact that people that don't natively speak English are much more likely to filter things through ChatGPT to "sound professional". I want your bad English. I want your bad art. I want to see the raw unfiltered expressions of humanity. I want to see your soul in action. I want to communicate with you, not a simulacrum that stochastically behaves like you would by accident. And if I want to use an LLM, I'll use an LLM. Now go away with your sparkle emoji buttons and stop changing their CSS class names so that my uBlock filters keep working. The human cost This year has been a year full of despair and hurt for me and those close to me. I'm currently afraid to travel to the country I have citizenship in because the border police are run under a regime that is dead set on either elimination or legislating us out of existence. In this age of generative AI, I just feel so replaceable at my dayjob. My main work product is writing text that convinces people to use globally distributed object storage in a market where people don't realize that's something they actually need. Sure, this means that my path forward is simple: show them what they're missing out on. But I am just so tired. I hate this feeling of utter replaceability because you can get 80% as good of a result that I can produce with a single invocation of OpenAI's Deep Research. Recently a decree came from above: our docs and blogposts need to be optimized for AI models as well as humans. I have domain expertise in generative AI, I know exactly how to write SEO tables and other things that the AI models can hook into seamlessly. The language that you have to use for that is nearly identical to what the cult leader used that one time I was roped into a cult. Is that really the future of marketing? Cult programming? I don't want this to be the case, but when you look out at everything out there, you can't help but see the signs. Aspirationally, I write for humans. Mostly I write for the version of myself that was struggling a decade ago, unable to get or retain employment. I create things to create the environment where there are more like me, and I can't do that if I'm selling to soulless automatons instead of humans. If the artificial intelligence tools were…well…intelligent, they should be able to derive meaning from unaltered writing instead of me having to change how I write to make them hook better into it. If the biggest thing they're sold for is summarizing text and they can't even do that without author cooperation, what are we doing as a society? Actually, what are we going to do when everyone that cares about the craft of software ages out, burns out, or escapes the industry because of the ownership class setting unrealistic expectations on people? Are the burnt out developers just going to stop teaching people the right ways to make software? Is society as a whole going to be right when they look back on the good old days and think that software used to be more reliable? The Butlerians had a point Frank Herbert's Dune world had superintelligent machines at one point. It led to a galactic war and humanity barely survived. As a result, all thinking machines were banned, humanity was set back technologically, and a rule was created: Thou shalt not make a machine in the likeness of a human mind. For a very long time, I thought this was very strange. After all, in a fantasy scifi world like Dune, thinking machines could automate so much toil that humans had to process. They had entire subspecies of humans that were functionally supercomputers with feelings that were used to calculate the impossibly complicated stellar draft equations so that faster-than-light travel didn't result in the ship zipping into a black hole, star, moon, asteroid, or planet. After seeing a lot of the impact across humanity in later 2024 and into 2025, I completely understand the point that Frank Herbert had. It makes me wish that I could leave this industry, but this is the only thing that pays enough for me to afford life in a world where my husband gets casually laid off after being at the same company for six and a half years because some number in a spreadsheet put him on the shitlist. Food and rent keeps going up here, but wages don't. I'm incredibly privileged to be able to work in this industry as it is (I make enough to survive, don't worry), but I'm afraid that we're rolling the ladder up behind us so that future generations won't be able to get off the ground. Maybe the problem isn't the AI tools, but the way they are deployed, who benefits from them, and what those benefits really are. Maybe the problem isn't the rampant scraping, but the culture of taking without giving anything back that ends up with groups providing critical infrastructure like FFmpeg, GNOME, Gitea, FreeBSD, NetBSD, and the United Nations having to resort to increasingly desperate measures to maintain uptime. Maybe the problem really is winner-take-all capitalism. The deployment of generative artificial intelligence tools has been a disaster for the human race. They have allowed a select few to gain "higher productivity"; but they have destabilized society, have made work transactional, have subjected artists to indignities, have led to widespread psychological suffering for the hackers that build the tools AI companies rely on, and inflict severe damage on the natural world. The continued development of this technology will worsen this situation. It will certainly subject human beings to greater indignities and inflict greater damage on the natural world, it will probably lead to greater social disruption and psychological suffering, and it may lead to increased physical suffering even in "advanced" countries. For other works in a similar vein, read these: Special thanks to the following people that read and reviewed this before release: Ti Zhang Annie Sexton Open Skies Nina Vyedin Eric Chlebek Ahroozle REDACTED Kronkleberry CELPHASE Share Facts and circumstances may have changed since publication. Please contact me before jumping to conclusions if something seems wrong or unclear. Tags: