I Don't Think AI Will Kill Junior Developers. But It Will Kill This Kind of Junior Developer.

I'm about to graduate, which is a very convenient time to be told that the career I've spent the last few years preparing for is apparently disappearing. Great timing.

Every other week I see some version of the same conversation. AI is replacing developers. Companies won't need junior engineers. Nobody should study computer science anymore. One engineer with agents will replace ten engineers. We're cooked.

Something is changing

I'm not going to sit here and pretend nothing is changing. It obviously is. I've changed. The way I write software today is completely different from how I wrote software when I started university, and I use AI constantly now.

When I look at some of the tasks that would've been given to a junior developer a few years ago, yeah, AI can do a lot of them, probably faster:

• Create a CRUD endpoint
• Write some unit tests
• Convert this component
• Fix a basic TypeScript error
• Write documentation
• Build a form
• Refactor this function
• Generate some SQL

These used to be useful ways for junior developers to contribute while slowly learning the larger system. Now I can give many of those tasks to an agent and get something back in minutes. That's real, and pretending it isn't happening doesn't help anybody.

It kills a version of the junior

But that's also why I don't think AI is going to kill junior developers. I think it's going to kill a specific version of the junior developer: the one whose entire value is “tell me exactly what code to write and I'll write it.”

Unfortunately, AI is extremely good at being told exactly what code to write. If that's the whole job, then yes, that's a pretty uncomfortable position to be in. But I don't think that's what being a good engineer actually is.

Writing code was never the whole job

That's where my perspective changed. For a while, I thought the way to become a better developer was basically to learn more frameworks, write more code, get faster and build more projects. Those things helped, obviously. But after building larger projects, working with other people and trying to turn random ideas into actual products, I realized that writing the code is just one part of the job. Sometimes not even the hardest part.

The hard part is:

• figuring out what needs to be built
• understanding a codebase you didn't create
• debugging something when the obvious solution didn't work
• reading documentation
• knowing when the AI is confidently wrong
• making trade-offs
• talking to somebody who isn't technical and figuring out what they actually need
• taking a vague problem and turning it into something concrete
• understanding enough infrastructure to figure out why your application works locally and explodes in production
• caring about the user instead of just the ticket

AI helps with all of those things. But it doesn't magically remove the need for them.

Move with the bar

Weirdly, I think that's the opportunity for people entering software right now. The bar is moving, so move with it. I don't want to compete with AI at typing boilerplate. That sounds miserable, and it will win. Instead, I've been trying to become the kind of developer who can use these tools to build more than I could've built before.

If an AI lets me implement something three times faster, cool. What am I doing with the saved time? Can I explore another approach? Can I test it properly? Can I understand the system better? Can I actually talk to the person using it? Can I build something I would've previously needed a five-person team to attempt?

That's where I think junior developers can become kind of dangerous now, because the definition of “junior” is getting weird. Imagine someone coming out of university who can build a frontend, understand a backend, set up a database, deploy an application, use cloud infrastructure, work with AI models, write agents, debug generated code, talk to users, and use AI to fill in gaps extremely quickly. Technically, they're still a junior engineer. But the amount of stuff one person like that can attempt is completely different from even a few years ago.

So that's how I'm trying to look at it. Not “how do I prove I can code better than an AI?” I can't, and that's a stupid competition. My competition is whether I can become better at building.

What actually scares me

I think universities are going to struggle with this for a little while too. If assignments only test whether you can manually produce some implementation that an AI can generate instantly, students are obviously going to use AI. You can ban it, put giant warnings in the syllabus, make everybody swear on their laptop that they won't open ChatGPT. They're still going to use it. The interesting question becomes: how do we make sure people still understand what they're generating?

Because that's the part that actually scares me. Not somebody using AI, but someone becoming dependent on AI before they ever developed the ability to reason without it.

I've caught myself doing this too. Something breaks and my first instinct is immediately: copy error, paste error, ask AI. Sometimes I'll deliberately stop myself. Read the error. Follow the stack trace. Figure it out. Not because asking AI is cheating, I don't care about that, but because I don't want the tool to replace the part of my brain that should be developing judgment.

I want AI to multiply my ability, not become my ability. That's a distinction I'm trying to be really intentional about as I enter the industry.

So no, we're not cooked

Yes, getting your first software job might become harder. Companies can produce more with smaller teams, expectations are increasing, and the definition of entry-level is getting blurry. Those are real concerns.

But software isn't becoming less important. We're building more of it. AI itself is software. Agents need infrastructure. AI products need interfaces. Businesses still have terrible internal systems. Security is becoming more complicated. Cloud infrastructure isn't disappearing. People still have problems, and most of those problems are not solved by opening a chatbot. Someone still has to build the systems around all of this.

So no, I don't think we're cooked. At least I hope not. I think we're just graduating into a version of software engineering where knowing how to write code isn't enough by itself anymore.

And honestly? Maybe it never was. AI just made that impossible to ignore.