Junior Valley of Death

Many LLMs and a dream – navigating a career shift and the Junior valley of death “Just ask the Chat” – navigating the Junior Developer valley of death

If you are a newly minted junior developer and are trying to land your first job, chances are you have come across uplifting articles stating that “The Junior Developer Job Is Dead. The Industry Killed It on Purpose.” and that you’re probably better off looking elsewhere. The generalized AI euphoria and business outcome min-maxing are having a serious impact on the junior job market because these tools can pretty solve less complex tasks a junior developer can tackle for much less—blocking junior developers from finding jobs in the software industry  and enforcing a pivot to other areas. In fact, even experienced developers are suffering from the pressure and we’re seeing news of massive layoffs at specific companies.

On the other hand, senior developers, which can further enhance their work output are also dealing with change, pressure and trying to adapt in their own way to the paradigm shift, which takes its toll on the availability to sit down with juniors and “teach the craft”.

While indeed there is a global shift in behaviour and how things are being done—in the sense that there is a lot more temptation to simply delegate tasks to an AI agent—, I don’t believe that there is reason for anyone, senior or junior, to panic. The key here is, as it always has been, adaptability and navigating the paradigm shift. While this may indeed be a challenging time for junior developers, just to keep up with the rhythm at which new things come out every day, it’s also a wonderful time where you can do a lot more with a lot more with a lot less formal training.

The job is changing. That, I think, we can all agree on. If you expect things to remain the same, and if you want to go about your business like AI doesn’t exist, then I’ll concede: that won’t work out. But sticking to a specific paradigm for your entire career has never really been the job, has it? 

There is generalized pressure for everyone to become a 10x developer right from day 0 now, especially in the “agentic age” where you can spin dozens of agents to tackle a wide array of challenges simultaneously. This is where things may get scary for a junior. Senior developers are swamped, already know the patterns and tasks like the back of their hands and having to sit down with a junior and go through the same flow for the nth time can be exhausting so it becomes tempting to say “just ask Claude”. Evidently, this can be a bit frustrating when trying to learn from the “sensei” you admire.

The project I’m working on right now pretty much needs us to condense what would usually be a couple of months’ work into a couple of weeks. Everyone is tackling multiple tasks (and even projects at the same time). It would’ve been impossible to do a year ago.

Now, it feels a little daunting, and it must be handled well to avoid consequences.

However, it can be done. We now have the tools to do it. I find that “use the tools available to you” may feel like brash advice, but it isn’t bad advice. The beautiful thing about these shiny new AI tools is that, with the right amount of patience, critical thinking, curiosity and adaptability you can have your own “pocket Senior” that allows you to debate and polish questions and further refine your ask so you can be more autonomous and have more meaningful conversations and questions to pose to senior developers—in a sort of “unsupervised learning” kind of way.

This makes up for a much faster, more productive and enhanced learning process where you are able to get the essential experience with less. With such vast amounts of information at your disposal and with integrated web search you can pretty much write anything anywhere in any programming language. Being curious and having high agency are essential traits—an AI agent is infinitely patient and you can question the reasoning and what’s happening between every semi-collon, which will allow for some beautiful discussions and rabbit holes where your knowledge keeps on growing and you start viewing more and more patterns and understanding them. This can happen even if you start with pseudo-code.

This, of course, is not always as straightforward as I’m making it out to be. The culture and environment around you also play a crucial role in this. This may come up as yet another career shift story but I’m confident that this is a case where AI agents enabled me to finally embrace my dream career as a software developer. Rewinding 20 years, I always wanted to work as a developer. Due to fate (or cognitive dissonance), I’ve ended up graduating with a Biochemistry degree and pursuing careers in technology and knowledge transfer, project management and ultimately business development in tech (having experimented with e-commerce along the way). 

Through the course of these careers I’ve always drifted towards tech and completed multiple programming courses on the side, always with a full-time job that made it hard to keep all the plates spinning. A career conversion was not on the table because there were bills to pay.

Enter 2023 and my previous job as a business developer in healthtech, which gave me the opportunity to work closely with the software industry. This was the final push to enroll on the master’s degree in bioinformatics that had an intensive introduction to programming, software architecture and data science. I felt right at home. And this is where the AI tutors truly shined, with very limited time to spend on the classroom due to a full time job, I had the opportunity to spend hours asking the machine about the concepts behind software development. I leveraged my project management experience and role-played quite a bit. I asked for the AI to act as the client and make a product request with requirements and acceptance criteria. I delivered those micro-projects for evaluation, whipped up an agent that acted as a senior developer that would evaluate code quality first, etc. This allowed me to continuously grow my knowledge and practice. After the first year, I was starting to feel confident enough to start looking for a new job in the field, which is where Subvisual comes in.

Subvisual has this sublime programme that is the Apprenticeships Programme, essentially a laboratory for growth as a software developer. If you’re selected, you get to work directly on a project (not an actual project for a client, something specific to you, so the pressure’s off on both sides—it’s just you and your will to learn), a blank canvas and an array of extremely talented and experienced individuals that are eager to share their experience and guidance.

This fits perfectly with my way of operating—having problems to solve with learning involved. So I finally had a real project, to fully implement Subvisual’s shiny new Content Hub from A to Z. A project with real requirements that involved working with real people and working with real senior developers that could chime in with real-world cases and experiences. This combined with an AI tutor that allows you to clarify all the basic questions and leave the really hard ones for when you need to bring in the “big guns” and learning “on the job” is like having a super power.

Fast forward 2 months and I’ve had the opportunity to integrate Subvisual’s roster of developers as a junior developer and day-by-day get to learn with a team of exquisitely talented engineers and experience a culture of openness towards exploring new tools and technologies and the vision that replacing junior developers (or simply replacing developers) with AI is ‘one of the dumbest ideas’ and instead viewing the A in AI as “augmented”. 

However there is indeed a “darker” side to all these tools that I experienced first-hand, which is the temptation and convenience. A big portion (if not the vast majority) of my colleagues on the master’s degree were completely hooked to the “Just ask Chat” virus, blindly accepting the outcomes without any questioning. I remember, when doing group work, spending a couple of hours peer-reviewing and bug fixing AI-written code and receiving the “Chat gave me that answer” feedback when asking for context. Even the teachers on the course spent less time and were less available and prone to throw a “Just ask Chat” when asked more generic questions.

This is where having a critical mind and being observant of the trend is important – the key now is curiosity and asking questions. The question mark should be prevalent on your discussions, be them with Chat or a human. If you learn to develop a strong sense of curiosity, in time, you will see that Chat is not coming for your job, he is the cape that you will be wearing with this new Super Power.

Evidently, none of this would’ve been possible without the right company culture and openness that Subvisual has – having true apprenticeships where you are learning on the job but without the pressure to deliver because you are actually working on a real project for a real client and what you are developing will be used. And most of all having a team that wants to transfer their knowledge and experience, teach you and learn with you in the process and not just forward you to Chat and move on with their lives. I believe this is the key to a successful and thriving industry.

I’m well aware that reality is not as sugary as I’m describing it here and this is a one-in-not-so-many case and that massive layoffs and unwillingness to spend the effort (and money) on hiring and training juniors are real, but I’m a natural optimist. In fact, a lot of companies are finally understanding the opportunity cost of not hiring people and training them in an “AI-native” way or even realizing that the AI isn’t actually paying off in the way companies anticipated or simply backfiring. Of course the pressure to deliver is still there and the speed is much greater and the dangers of context-switching fatigue, cognitive-overload and AI-brain-fry are real since you can now deliver a months-worth of work in a week, and there’s pressure to learn faster than before. 

However, junior developers always had an expected timeframe to unravel, with the respective underlying pressure (albeit self-applied more often than not) to deliver and be productive on an already fast and competitive industry – AI just cranked the speed to 2x (just the way we like to listen to our podcasts).