What next for junior developers?

Everyone is worried about junior developers. What are all these fresh-faced computer science graduates going to do now that AI is writing all the code?  

It is a legitimate concern. 

It wasn’t that long ago that the best advice I could give an early-career person interested in software development was to go to a boot camp. Sure, they could go to college and get a four-year computer science degree, but that would be expensive, take a long time, and teach them a lot of theoretical but impractical things about computers. And they wouldn’t even be doing science. 

But a six-month boot camp? There they’d learn what they really need to know—what software development companies are really looking for. They’d learn practical coding techniques, proper bug management, design specifications, JavaScript and TypeScript, source control management, and continuous integration.  

When I was a hiring manager, it didn’t take long for me to realize that a boot camp graduate was often much more ready to hit the ground running as a junior developer than a computer science graduate. 

But of course, all that fell apart overnight. Suddenly, for a low monthly payment, I could have a tireless, diligent, eager, and highly skilled junior developer who can type a thousand words a minute and reason at the speed of light. The economics of that are simply too compelling. 

Juniors begat seniors

And so what is a budding software developer to do? Or more importantly, what is a software development company to do when they realize that all those senior developers who are using Cursor are actually going to retire one day?  

Up until about 10 minutes ago, those companies would hire these intrepid young whippersnappers and put them to work fixing bugs, writing the boring code that builds systems, and slowly but surely teaching them how systems work by having them learn by doing. One became a senior developer through the experience of writing code, seeing it run, and learning what works and what doesn’t. Eventually, wisdom would set in, and they’d become sage, seasoned developers ready to mentor the next generation of developers.  

Well, we are now skipping that part where you actually become wise. But wisdom is actually the critical thing in this grand process. The judgment to know what is good, what is effective, and what is needed is the very commodity that makes agentic coding work. The AI model writes the code, and we seasoned veterans determine if it is right or not. 

We seasoned veterans know if the code is right or not because we’ve written tons and tons of code. But humans aren’t writing tons and tons of code anymore. And here is where I’m going to say something that I think many of you will really not like: Code doesn’t matter anymore. 

What I mean is, code is a commodity now. Code that used to take months to produce can now be produced in minutes. Yes, literally minutes. And the coding agents today are the worst they will ever be. They are only getting better, and they will only produce cleaner and cleaner code as time marches on. At some point—and that point may already be here for many of you—we are just going to stop looking at code. 

What matters is whether or not the application, you know, actually works. And if you want Claude Code or Codex to write a working application for you, you need to be able to communicate with it effectively to get it to do what you want. And strangely, the way to communicate with it is to write clearly. 

Heads up, English majors

A couple of weeks ago, I wrote that Markdown is the new programming language, and that what makes for “good code” in Markdown is the ability to write clear and concise instructions. Who would have thought that the English department would suddenly be the key to developing good software? 

Right now, the agentic coding process goes something like: 

1. Describe the problem to Claude Code.
2. Monitor the code Claude writes to make sure it is good code.
3. Test the application to make sure it works correctly.
4. Refine and improve by iterating this process. 

Step 2? It’s already becoming unnecessary. These AI agents are already writing good code, and the code they write gets better and better every day. And it is trivial to tell them to improve the code that they have already written. Iterating to improve code quality takes mere minutes. Writing the code has literally become the easiest part of developing software. 

So my advice to the kids these days: Learn to write clearly and precisely. Learn how to understand systems and describe them and their use cases. Make sure you can succinctly describe what you need software to do. English majors take note. Hiring managers? You too.