待翻译：John Henry in the Age of AI

AI 服务暂时不可用，以下为来源正文，待恢复后补全翻译。

John Henry in the age of AI 2026-08-06Dave #ai #programming There is an American folk tale about a man named John Henry. It dates to the mid 19th century, a time during which America was rapidly industrializing and the use of steam power was picking up… Well, steam. Being a folklore tale, there are many different variations of it, though, most examples describe him as a man employed on the construction of railroads. His specific role is often described as a “steel-driving” man but this takes various specific forms such as hammering steel drills into rock to dig railroad tunnels, or hammering the spikes that hold the rails to the wooden sleepers. Generally speaking, John Henry is depicted as an extremely skilled and well respected worker. There are many full-text versions available online, but I’ll specifically reference this one. The synopsis of the tale of John Henry is as follows: John Henry is a railway construction worker and sees a steam-powered machine performing his job at an incredible pace. He decides to measure his prowess as a railway construction worker by having a duel against the steam-powered machine. The duel is usually something along the lines of racing to mine a railway tunnel or set and spike down a length of track, with the winner being the fastest to complete. John Henry ultimately beats the machine, but the exertion he put his body through to do so causes it to give out, and he passes away. It is the classical man-versus-machine tale. And I think the parallels that can be drawn between it and the emergence of AI in recent years are clear, and I will discuss them later on. I heard the story of John Henry when I was a child, but I forgot all about it until just recently. I subscribe to several different interpretations of the folklore tale, but, in my opinion, it ultimately boils down to a statement on how ever-advancing technology disrupts and conflicts with human dignity, perseverance, ability, and purpose. It is a reasonable conclusion that John Henry felt threatened by the emergence of technology that could replace him. And his response of setting up a duel against said technology suggests that he wanted to prove that he, a human, can equal or defeat the machine. And the outcome of the duel asserts that man, the human spirit, ultimately triumphs over machine, but at a cost. This time, the cost was John Henry’s life. Looking back on this story almost two centuries later, with a modern perspective over the technology we have now, coaxes out some obviosities. The use of machinery in tunneling and railway construction are the norm. Construction without machinery and with only human labor is practically non-existent anywhere machinery is available. There are still “steel-driving” men, but their tools are no longer hand tools like John Henry’s, but machines. And the benefits of this shift to mechanization are obvious: such projects are completed faster, consuming fewer resources, and - most importantly - more safely and with a lower human cost. Other external factors like safety standards evolving over time too absolutely come into play here. However, having fewer humans on the job site, and more machines, ultimately expose less humans to harm for simple mathematical reasons. I assert that the meaning behind the tale of John Henry is directly applicable to what is happening right now. It can easily be interpreted as an allegory describing the emergence of AI and the impact that AI’s emergence has on how labor is executed and what the value of a human being is with respect to work. Let’s interpret the tale of John Henry in the context of the last couple of years of technological advancement and the emergence and adoption of - and opposition to - AI. I’m a software engineer, and will largely speak from the perspective of my career. Software engineering is a field some speculate will be replaced by AI. Boris Cherny is the creator and head of Claude Code at Anthropic and, Anthropic is currently one of the top companies that actually creates AI, rather than consuming it. Cherny famously stated that “coding is largely solved” - to the chagrin of many software engineers everywhere. That’s our first parallel to John Henry. A machine has been created that, in theory, can take the place of a human and do a task a human was previously required for. This corresponds to the steam-powered machine John Henry dueled against. Sidebar: I completely understand that you can’t remove a human from their desk and plop an AI there and get 1:1 results. I am not making the argument that AI is sophisticated and accurate enough that this is doable. I write code for a living. I have seen the code that AI generates. They are not on the same level yet. Yet. I am also not commenting on whether I think AI will ever reach that point. It is inarguable that technology advances and gets better, but it is not a given that all technologies advance and get better. Plenty hit their limit and do not evolve further. Let’s get back to our steam and silicon powered machines.1 Watching the emergence of AI, I have absolutely felt the threat of it, personally. Watching a machine in seconds churn out a Python script that would have taken me at least an hour to write is frightening. Why pay me to spend an hour doing that when a machine can do it instantly? It felt like looking at my replacement. And I can imagine that John Henry, nearly two centuries ago, felt the same dread looking at the steam-powered machine doing his job. How would my story look if I decided to duel the machine replacing me, like John Henry dueled his? I’ve been programming since the mid 2000s, and professionally since 2010. I have always followed emerging technology, including AI, and looked for ways that I could leverage them to be a better developer. By now, I have a decent amount of experience using AI tools to execute my craft. That being said, I look at the output of AI tools regularly and I have observed that it generally does not compare to code written by skilled humans. But that is not too surprising. Communicating the requirements of a software project often trips up humans too. There is an entire field of work dedicated to communicating and tracking requirements - project management. So it comes as no surprise that a bot lacking context outputs code that misses the mark. On top of that, AI is technologically limited by constraints like context windows that can easily lead to similar breakdowns. But I also recognize that the code LLMs output is often “good enough”. It completes the task, but is written with patterns I don’t agree with or know better approaches to. And sometimes, that’s okay. The output can serve as a general skeleton, ready for improvements by humans. It can make assumptions that even a human missed, and now needs to consider. Or if the task is trivial enough, say one-off tasks, “good enough” might just be good enough. I also need to mention that sometimes AI writes code exactly as I would have written it, which feels amazing and magical. Quality does matter to me. I cannot stress enough how much it pains me to ship ugly, but working, code. Even outside the context of AI. I fully recognize Steve Jobs to be a deeply flawed man, but I adore the story of his interaction with his father in which his father explains the value of a job done right and with care. The story of John Henry doesn’t touch on the quality of the work or whether it mattered to him. But, his position on the matter can be inferred. Versions of the tale include: He picked up a hammer and a little piece of steel Saying, “Hammer’s going to be the death of me, Lord, Lord, Hammer’s going to be the death of me.” John Henry dedicated his life to his craft. And that brings us to the crux of what I’m trying to convey: He cared about the craft and pride of the work itself, not just the outcome. And the same is true when it comes to AI: If you care about making and doing, and the process is more important than the result, AI isn’t for you. If you’re trying to get a job done, AI can be a force multiplier. The code might not be great, but it just might work. Human code needs review and tweaking, and that doesn’t go away with AI. Human-written code can still be slop and that slop can get sent to production if it goes unreviewed. These are the two extremes. The far, far ends of the spectrum of AI use. What defines the reasonable middle ground? Quality defines the middle ground. A skilled developer can use AI as a force multiplier to write more code at the same quality he would have otherwise, but faster. And that’s what I think a major part of the debate around AI is. What I’ve revealed here is that it all boils down to what one’s viewpoint of the purpose of a human being is. Everybody has to work to make the world and society go around. A lot of people think this sucks, and working your entire life is a tragedy. And I’m inclined to agree. But I also see technology as an escape from this existence. In John Henry’s time, digging a tunnel required the toil of thousands of men and women. Or, similarly, farming required hundreds of men or women working the fields, becoming sunburnt and ruining their backs bending down to reach the plants on the ground. These occupations were dangerous and took an extreme toll on the body. But those jobs are done differently now. Farming no longer requires thousands tending the fields year-round, almost all of the work can be done by machine, with one machine replacing many, many workers. Ditto for tunneling. The men and women that would have been dying or destroying their bodies doing such labor are now doing something else. That something else might not take a similar toll on their body. And that’s a good thing. I’m describing a net reduction human suffering. And I’m describing manpower that has been freed up that can now be put towards something else that improves humanity, like figuring out how to replace jobs that are necessary but incur human suffering with solutions and machines that don’t suffer. However, it’s worth mentioning that shift doesn’t happen overnight. People out of work because they were replaced by automation can have a hard time finding a new job. Starting over and shifting to a new field of work is often necessary due to the very automation making the job market in a specific field disappear. Some people become poorer. There are immediate and real world consequences worth considering. This is a legitimate grievance. But, machines replacing humans isn’t inherently a bad thing. It can be beneficial as it frees up labor resources - those humans - and potentially allocates them to fields that advances the knowledge and capabilities and comfort of the human race. And it’s a snowball effect. The observation that technology is advancing at a faster and faster pace is a common one, and accurate in my opinion. I dream that, one day, machines are intelligent and accurate enough that mankind is freed from labor entirely. That one day we’ll hit a point of technological advancement that allows most, if not every, necessary function in society and the world to be handled by machines that don’t need to be paid and don’t suffer. And, unlike the above descriptions where this now-freed up manpower can be redirected into some other form of labor, it can now be directed into leisure. I feel that I need to mention that I believe some roles will probably still need a human touch forever. Social issues, perhaps. The non-financial cost of AI is a hot topic and worth connecting to John Henry as well. It has been said that the resources consumed by AI damage the pl [truncated for AI cost control]