The 4th Industrial Revolution: How Everything Will Be Automated

Fire, steel, electricity, data and now? Intelligence… The media is talking about a revolution, something more disruptive than anything that came before. Intelligence is coming for our jobs, minds and lives. Is it true? I was wanting to write a blog about this for a while, however, I always felt like I do not have my arguments ready. However, I gave myself a rule that I should be pushing out posts even though they are not perfect, so here we go!

In this post, I would like to pose my thoughts in 3 paragraphs: (1) where do we already see the impacts, and what can we learn from them, (2) what does this mean for jobs and (3) what shall we be doing as individuals to prepare for the future.

Software Engineering Psychosis

When I was choosing a career, I was fascinated by artificial intelligence. Back then it was not exactly what everybody was doing, but rather a niche technical track that was quickly gaining popularity. Since then a lot has changed. More or less successfully I have crawled myself towards the end of the Master’s studies (not done yet!). However, I realized, I did not know what kind of skills or vision I had entering the work force. Therefore I have decided to follow my passion and started an internship, and this is where a massive change has come.

At my work, I was suggested to use Claude Code. I knew about the tool, but did not have the access to the full paid version before and I have quickly started using it. Claude Code is a nice little AI, that sits at your computer, looks at your files and helps you programming. However, over the last year, it has evolved from something that helped you find a typo or format a file to a beast that can plan and execute 1-2 hours of coding efforts. As I write this I have Claude trying to fix something for me, it works completely autonomously: it plans, sees its errors and proceeds to solve them.

Yes, it makes mistakes. Yes, it is not perfect. Yes, it did spawn so many containers that it killed my memory and I had to restart my laptop. Yes, and when I told it next time it pruned the containers such that it killed processes that were vital for other things on my laptop. But yes, it does mostly work. And it is incredible, it is better the longer you use it and it does everything you have ever dreamed of as a programmer.

Suddenly, the time spent at work is not the bound, the bound is our capacity to the human brain that can coordinate processes. People are stopping to read the code that they write. The best example of this is Peter Steinberger, creator of OpenClaw (AI that takes control of your PC), which famously coordinates fleets of these autonomous coding agents.

Incredible implication of this is that you can define a problem, a metric and constraints, and let agents just work on it. This goes by the term autoresearch, term coined by Karpathy automates training of large language models using that. Does it work? Yes, it does, but no better than previous tools. More than actual results, it is important to ponder the power of this concept. Autoresearch is so much easier to setup. This is not some crazy math optimization problem somebody tuned for ages, this works like a human would (also with its errors and drawbacks) and it can iterate whole night while you sleep. Only cost? Tokens, and grounding of the experiments (e.g. in the lab or by computing the results on a GPU). Incredible.

Jobs Will (Not) Disappear

Okay, okay, okay, Michal, what does this mean? Are like all programmers replaced? What is the next job AI will take? I think the reasonable thing to do is to look at what has happened in the past. I might be very much influenced by the recent Karpathy’s interview, in which he brings up the example of bank tellers. The invention of ATM has not killed the job, rather the contrary, the access to money has become seamless, and suddenly there were more jobs in the industry. Only later has the number of people employed in the area decreased. This means that the jobs will not disappear, but rather the entry bar will be lower, and there will be more (also less qualified) jobs in the area. Now you do not need a Master’s degree to be bringing up a website, basically anybody could do that. The highly qualified jobs will get even more interesting, having the right ideas and understanding the right concepts – creating something people actually want will be even more important.

Currently, the capabilities of tools like Claude code have increased massively, giving developers wings. I can tell you that from experience, cause I always had problems with exactly the aforementioned: websites. Figuring out ports, backends and frontends, never wrapped my head around it. Now I just say, claude "do me a little website" and it spins it up in a few seconds. That being said, I still struggle with usage of these tools. Especially starting a new work, I have realized that I am extremely disorganized, both in my ideas and my execution. The fact that Claude accelerates my development does not mean I produce a better product, I actually just generate garbage faster. Additionally, I am slow in managing my agents. I cannot do more than two instances at a time, and my context switching has a huge cost. I am not able to read anything properly and process it, when I am waiting for my agent to finish, it is just shallow processing. I have to learn to use these tools effectively: plan, execute, with right permissions at a right time, delegate well-defined tasks, and understand the outcomes. Document and test prototypes, even more than before, I have been given a chainsaw to cut bread and now there is a lot of mess around.

Also, we really do not know what is going to happen. We are fundamentally bad at making predictions. I was just having a blast reading the The Fourth Industrial Revolution by Klaus Schwab. It is written in 2016, and it talks about the revolution that is coming. You might ask, how did he know ChatGPT was coming? Well, he did not. He predicts that by 2025 we will have small cities with fully self-driving traffic and houses being printed by 3D printers. He is not alone in thinking this, he presents statistics showing a significant percentage of experts being certain in this. And they are also not completely wrong, those technologies have indeed developed, however, different technologies, which turn out to be even more important are not even mentioned (most of the ideas of the book still apply though). This means that fundamentally, we have no idea what is going to happen. Thinking about what jobs will be there, and which not, makes sense, but we should not be worried.

One point of the book that I really enjoyed was that big revenues are made with smaller and smaller amount of people. Back in the age of cars coming out, you needed a whole city to run a globally successful company. If you wanted or not, you had to give jobs to people around, and build up the ecosystem, and if nothing else, you needed their labour. This is not true anymore now, the same amount of revenue is made by smaller and smaller groups of people, be it Tech giants, or the rising AI Labs. This is precarious, given that intelligence is computational (more on that in this post about ‘What is Life?’ by Blaise Aguera y Arcas), meaning that whoever controls the tokens is gonna control intelligence. The tokens / or whatever unit we are going to have is the raw power behind the future revolution. If the trend continues, it is plausible, that human power will not be needed for economic activity anymore. If this happens, this might be a problem for human well-being. Now states are motivated to keep us alive and happy, cause we bring them products. If human existence is automated, then why would they care? The arguments for this are nicely summarized in Curse of intelligence. In this same source it is mentioned that paths to prestige are changing faster than ever, cause the world is changing so much. I like this, cause this means you might just as well think for yourself instead of sinking time to please a system that might not even work in the end. With all the startups it seems that taking chances and bets is rewarded more than being hard-working and consistent.

Thriving During a Revolution

I have recently joined a machine learning research company, and that has pushed me onto the edge of the actual development of what is actually happening, especially on the tools that we are using and the channels that we are following. This has made me feel the shift even more. The last week I took a step back, I stopped burning so many tokens on Claude code, and started writing the parts of code that I care about myself. I feel like writing code makes me more aware of what I do, and I think more clearly. I would like to combine the clarity of thinking with speed of Claude, but I am not there yet. Either way, it made me rethink how I approached work with LLMs, and made me realize that the capabilities are limited by me, not by the tools. My understanding, clarity of mind and communication with others about my goals is what limits me. For myself, I have made a little guide of how I want to treat this. In my blog a year ago, I famously said I stopped using ChatGPT, and it got controversial among my friends. There is certainly an unsolved question of what is the best way to collaborate with these tools. My message this time, however, is that there is nothing to fear, and all is going to stabilize (eventually). Meanwhile, I wrote a couple of points for myself of how to sleep calmly when it is storm outside:

1. _Read science fiction._ I am a firm believer that our limit is our imagination. You cannot live a life that you cannot imagine, may that be a profession, relationship or a company. It always starts with your subconscious. Start dreaming. But also be careful what you pray for. Either way, especially in times of AI thinking for you it is nice to exercise your own imagination. Try to dream up a future, and you will most likely be wrong, but that is fine. The world of Dune or Contact is now closer to reality than ever.

2. _Curiously learn about the tools._ Tools are here to stay, so regardless of what it does to your job, you might just as well take advantage of them. Learn to use them, and use them well, such that they do not disempower you, but rather lead you to deeper insights.

3. _Create your own things._ Nobody cares about AI generated music. No, really, not. If you want to create something of value, you have to actually put in the work. AI does not have a hunch, and does not understand a product market fit. Good innovation is still precious.

4. _Protect your mental space._ Be conscious about what you are doing with AI, and where it takes over. The stuff you one-shot overnight might not be so valuable in the end, and maybe one day of actually thinking is more valuable than one day of vibe coding, even if you feel like you have magical amount of progress. Francois Chollet says that the time to learn how to think for yourself was before genAI, if you missed your chance, good luck, but I do not think so. It is just something that we have to be more conscious of now, same as physical exercise is something encouraged to prevent obesity.

5. _Believe._ Does not have to be religious. World will not be the same, nobody can guarantee that. Or more so, I can guarantee it will not be the same in a year. If you start studies now, you will finish them and the world will be different, maybe your job will not be needed anymore (just like it seems to be happening to mine). It does not make sense to try to study for a job.

Good, now that I have spent the sunny Sunday locked in my bedroom typing on my laptop, I can leave this blog post and go talk to my Claude as we finish together my other projects. And on Monday I will get up and try to advance intelligence further.

It is incredibly exciting times. And oh yeah, this text is fully written by me. Human tokens. Awesome, right? (Now that I think about it I just created more training data for AI.) Thanks for reading so far, really appreciate it.