System Failure: The Great Convergence and the End of the Junior Role
I’ve been listening to Nate Jones on YouTube-that’s the guy who runs a really successful AI Substack-and I watched a couple of videos of his. I have to admit that he changed a couple of things about my understanding of the current AI landscape. Because obviously, as a product person, I’m not coding myself, I’m not using Cursor on a daily basis. Yeah, I’m using Claude Code now, and I’m using a lot of Lovable and so on.
In the first video that I watched a couple of days ago, he claimed that there is this converging career path now. Because previously, obviously, you had the developer, you had the leaders, something in between, you had designers, you had product people, you had analysts, and so on. But at this point of time, basically, product people started to use Lovable to build things. That means they don’t need designers anymore to such an extent; they don’t really need engineers to build prototypes and simple things. At this point of time, the very same goes for engineers, and the very same goes for designers.
I personally know one of the designers at my current gigs-one of the projects that I worked on-and this designer is basically not designing anymore. He is building prototypes, and he’s pretty well-versed in hosting this and building backend even in Claude Code now-a really simple one, hooking it up with Supabase and these kind of technologies.
And he’s right. Nate is right. There is this big elimination of the white-collar work. These things are being merged; there is no career path anymore. If previously you would be using something like roadmap.sh, kind of planning your roadmap, your career, like five, ten years from now with steps that you have to do in the middle… basically the whole thing is collapsing. Collapsing in a way, more on the implosion side, right? So now, the whole thing, the whole career path is like five months. So the time frame for the change-so time to change really-is just dramatic. And there are no standards, no workflows. The security is obviously lagging behind anything that Anthropic, OpenAI, and Google do really-Anthropic being the best among them.
And clearly winning the race. This means that if you are not on the bandwagon right now, you are not working on your skills, you are not playing with those tools… because the analogy that Nate is presenting is basically you riding a bicycle, right? So the faster you go, the more stable you get. And he makes the case that if people are sitting this out, trying to wait until there are some kind of a guidebook, laws, standards for the PM to use the AI, this won’t really exist.
So the question now is whether you get on the other side, you get on this train, and you build your own workflows, your own understanding how AI is going to be the engine that sits on top of your domain knowledge, of your experience, your ideas. And if you’re able to incorporate it now, you just have like maybe seven, eight months, maybe a year or so, just to figure out how the whole thing will incorporate your specific domain skills. Obviously, the domain skills are not going to disappear, which means my code will be of worse quality based on my instructions, obviously, and the application business-wise might be worse on the end of some engineer who decided to build the app himself, 100%. The same goes for designers. Even so, frankly speaking, if they are building something from scratch, they will be really good at UX, applying the modern design practices, and pretty much copying the interfaces, but they might have issues-same with me-like the backend things that might be not that secure and so on. And obviously, some assistance on the senior side will be needed.
Having said that, obviously, the domain knowledge is not disappearing, but AI probably eliminates the entire entry point for juniors. For the time people and companies would invest into them to develop those skills to become regulars, to become seniors. Seniors will be using the AI, and the juniors… I’m not sure about this, if companies will be ready to invest that much money into juniors, unless you as a junior invest a lot of time yourself just to educate yourself, not only on the fundamentals but also on the AI usage. And even in this case, you won’t be able to compete. Because in this case, if the Claude license, or Gemini license, or OpenAI license costs that much, then companies can provide them with skills, with guidelines, with clear structured projects and so on. So in this case, the experience of the engineer is not that important anymore. They can hire someone from a developing country, and they wouldn’t really care that much about this because they have standards and the AI agents are getting better.
Just to underline this point: Claude Opus, the new one, basically they experimented with this running for two weeks, and it wrote the entire C compiler, with 160,000 lines of code, I think. Having this whole thing in memory, testing it out, and it’s perfectly functional and so on. And it ran independently for two weeks. Which… well, it is pretty much a regular developer at this point of time, not even a junior anymore. Obviously, if you are working with a legacy codebase written by humans and so on, this only means that the seniors are still there to prepare the whole setup, the codebase behind it, just to make it usable/operational on the AI level.
So soon enough we’ll be custodians, butlers, cleaners, people who are just there to keep an eye on AI agents, give them orders, controlling the outputs and so on. So we are the operators of the machines at this point of time, until they are good enough with reward functions to build those things themselves really, and even testing in production. You can imagine a future where we have the agent writing different types of functionalities, testing them out with different audiences dynamically, and figuring out which version performs better with a certain reward function that would be provided by the business owners. Or not even them, because the AI would have those reward functions and proposing them, or even testing them out, but obviously it would need to understand the current KPIs and the company. If you happen to have any at this point of time, because there is this degradation of skills on the business side, on the design side, and pretty much across the entire Western world, I would say. So I’m not sure how this develops.
Having said that, there is one note here. I think as PMs, designers, developers, we only have about maybe 7, 12, 24 months before those tokens become really expensive. So everyone will be really locked in terms of vendors and hooked up on the system hopelessly. Which means that companies like Anthropic will be able to charge you a really solid piece of money. And soon enough, well, these companies now are buying those licenses for engineers and pretty much everyone who wants to play around with this. But soon enough, they will be so expensive that only the seniors, the senior staff who know what tokens are, how expensive the whole thing is, how to engineer the context-not just prompting, right, but the context engineering, the dynamic things just to explain to the agents how it is supposed to be acting, preparing the codebase, the instructions and so on-only the senior guys who will explore the topic to the best of their ability, they will know how to oversee those agents, how to order those pieces of works. Only these people will start getting licenses because they will be really expensive.
So soon enough, there will be this gap with people who invest a lot of their private time these days just to learn the systems, to learn the building. And everyone else who is just sitting it out and waiting, and waiting for some standards to come and for those models to become really cheap… I don’t believe they will be. At some point of time, we all expect the VC money to run out, and this means that these people will have a need to have the real economy-you need economics behind the product. This means the tokens will become really expensive. So soon enough, these companies will start saving and killing those licenses. So only key people who learn to use them now will get a chance to operate with AIs, and they will have this really immense, incomparable, revolutionary advantage because they would have the experience, they would have the knowledge, they would have the skills.
And obviously, a lot of people are writing crappy code or creating crappy code now in companies they work for, when the company is also paying for those licenses. But these people who are risking it now, they will be in order. They will have the skill to offer in their next gig, next job, next project where they will be able to use the tokens in a proper way, and they will be able to save those companies the money. There will be consultants who will prepare the companies with legacy codebases to make the whole thing more agent-ready and so on.
But I think that we don’t really have that much time as PMs, as designers, as engineers, especially if you are a junior or regular. You just have a couple of years probably to get ready to learn how to automate those things, how to steer the whole thing, how to learn building with AI. There is no way around it anymore. Yes, you can check out, you can switch your career, but if your intention is to stay in tech, there is no way around it really. Not anymore. So it is our way or the highway. And highway in this case being you switching to something not tech… not tech-heavy really, some handwork, some sales, whatever, I don’t know. Maybe some engineers will move out to be the developer experience advocates or the sales people, the account managers who will help out those AI companies to hook up those companies out there even heavier on AI. And these engineers will have the know-how, they will have the knowledge-well at least they will have enough technical knowledge to prepare the codebases and integrate companies even more with those AI agents of a given provider.
And at this point of time, I don’t think we are at the point where the LLMs become widely spread in terms of local models, something that you can host yourself. Not really. It just looks like that we will end up with another really crucial and big subscription. If it goes down, if you can’t really afford it anymore, you just lose like 80, 70 percent of your competitive advantage if you had any. And there is no way around it really. The swarms, for example, those groups of agents that you have on Claude Code at this point of time, it is only the beginning. And looks like the speed of change there, it speeds up the whole thing. It speeds up really. Things that we saw like half a year ago, they are just completely different today. And no one is up to date at this point of time. And you have to dedicate a lot of your own private time after work just to be relevant if you want to stay in this industry, if you want to reap those rewards, if you want to keep earning the bucks. You just have to stay. After work, you have to learn, and you have to squeeze every opportunity in your workplaces at this point of time to use AI in some capacity. Because if you don’t, there will be a lot of people expecting that there won’t be that many job places in the white-collar industry anymore. This means that only the seniors, only people with really heavy experiences and really, really know-how they possess in regards to AI and this integration of AI and their skill set-this is the only profile, career path now. It’s just for you to dive in with your own private funds or the funds of your company, or better the two together, just for you to learn the tooling and to integrate those into your workflows, whatever it is. Because we see lawyers using them, we see the medical using them… everyone is using the AI really these days.
I’m not saying the chatbots. Chatbots is pretty much the topic for the plebs, the lowest class really of the internet users. And the AI agents is something that all of us have to learn at some point of time. And the faster you start, the higher the chance is that you won’t be thrown overboard the moment someone comes with a better skillset. Because again, your advantage at this point of time is the domain knowledge. And the future AI users who might be better at orchestrating AI agents, they won’t possess this experience, this domain knowledge of something that used to be is not that easy to acquire. You can be really brilliant at AI at this point of time, but if you don’t have the experience like a decade or something under your belt, you’ll still have problems. And this is still a comparative advantage, but it will die off pretty soon.
So the convergence is expected. The convergence is happening already now. I’m building prototypes in Lovable myself, I’m learning Claude Code and many other things. There is no way around it, guys. If you want to stay relevant, this is the way to go. So this year, you can forget about many things, many hobbies and so on. You have to waste a lot of hours to get yourself up to speed or you’re just risking your future here. You are just betting that those things won’t pan out, they won’t work out, and something will collapse. But from the things that I’m seeing now, the LLMs are here to stay. There is no way around it. I was wrong. I was expecting those LLMs to reach some kind of a limit, but I think that we just didn’t really reach that point of time, and they are good enough at this point of time to build a lot of tools that we have. We are not that unique. We build the very same tools in different organizations that just do and happen to have the very same goals, the very same reward functions, and the very same outcomes that the business is expecting.
So, if you are not on that horse, not on that train, you are royally screwed.