Develop Yourself

#278 - AI Does NOT Replace Junior Developers, Here's Why

• Brian Jenney

Share episode

Every headline says AI is eating developer jobs. But spend a week in production and you’ll see the opposite: brittle code, flaky tests, and tools that look fast until you actually ship something.


In this episode, I break down why the “AI replaces engineers” story sells so well—to investors, to execs, and to lazy headlines—but falls apart in the real world. We’ll talk through a Cornell + METR study showing seasoned devs got 19% slower using AI (even though they thought they were faster), and why that tracks with what I’ve seen on real teams.

Send us a text

Shameless Plugs

Free 5 day email course to go from HTML to AI

Got a question you want answered on the pod? Drop it here

Apply for 1 of 12 spots at Parsity - Learn to build complex software, work with LLMs and launch your career.

AI Bootcamp (NEW) - for software developers who want to be the expert on their team when it comes to integrating AI into web applications.

SPEAKER_00:

Welcome to the Develop Yourself podcast, where we teach you everything you need to land your first job as a software developer by learning to develop yourself, your skills, your network, and more. I'm Brian, your host. The CEO of AWS recently said that the firing of junior developers because AI can replace them is quote unquote the dumbest thing I've ever heard. But is this even controversial anymore? Depending on who you listen to, you'd think that developers were basically already obsolete. Every week it feels like another company announces layoffs. Even companies reporting record profits like Cisco are using AI as the excuse to cut jobs. And I get it, but I don't like it. It's convenient, right? It's not like inflation or tariffs, interest rates have anything to do with this at all. But hey, what do I know? I'm just a lowly software developer. And we all know that CEOs of massive corporations are exceedingly trustworthy, right? Here's why people might say that AI is replacing jobs. Number one, the money's there. Billions of dollars, maybe a trillion dollars have poured into AI. Investors need a story. They really need you to believe that AI is replacing workers. And this is the easiest story to tell because everybody just eats it up wholesale. Number two, it's a good scapegoat, which is kind of related to number one. Executives can trim bloated orgs from pandemic overhiring and they can blame AI instead of their own bad planning. If you're part of a bigger organization, maybe you remember, maybe you still have this where you have a manager that manages another manager, and that manager manages three people. I actually was kind of one of these people, and I got laid off too. No hate, no shame, but you know, yeah, you're getting paid these outrageous salaries to really manage a team of a few people, or maybe you're a manager of a manager, of a manager. Maybe there is some bloat that needed to be trimmed. Number three, and this one really hurts to say developers have become unlikable. We just aren't that likable. For years we were posting day in the life stories, TikToks, cold brews, kombucha,$400,000 salaries to make a button with React. Is it any wonder why people do not like us? So when they hear that we're being replaced, they cheer. And can you really blame them? And number four, people might be saying AI is replacing jobs because it will. There's almost no doubt that there are some jobs that are going to be more at risk than others. Will it create as many or more jobs than it takes away? I don't know. And no one does. And anybody claiming to know that is just telling you a lie. I do know that the reality that we're living in right now is messier than what the headlines have said, and it's also a little bit more boring, too. And here's why I don't believe AI replaces even the worst of the juniors that I've worked with over the years. I use AI every day. I use Claude, Cursor, ChatGPT, starting to use Codec. Here's what people miss AI is not fully autonomous. And if it was, I'd be very, very nervous about using it anywhere. It can't join a team, it can't understand the context of the team dynamics, it can't sit in meetings, it can't look at designs, it can't make changes across multiple code bases and open up a pull request, accept review feedback for that pull request, and then deploy that code to production through separate stages and layers of checks and balances and things like that. That's a human loop, maybe forever. There was actually a big news article that came out today where I believe Deloitte is now having to pay back half a million or tons of money to some government because they essentially generated a report with AI that was incorrect, right? And they didn't have a human check it apparently, and it was just made-up stuff. Hallucinations happen. I don't care how good the model is, hallucinations are still a thing that we are dealing with. And if you have no human in the loop, well, you get to these really messy scenarios and situations where things just break. And who do you blame at that point? Anthropic, open AI, Grok, and then who foots the bill for these kind of things? So humans are going to need to be in the loop. The next reason I think we're going to need junior developers going forward is because the code that these tools write is excessive. If you ask an AI coding tool for like a helper for a function or something, you get these long-bloated code snippets. And if you ask it to build something as complex as a data pipeline, you can get something really brittle that if you don't take time to diagnose and fix and check, you can end up with a house of cards that you won't know how to fix. The more complex the ask, the more garbage comes back. And next up is this 80-20 myth, which I kind of believe is nonsense. People love to say AI does 80% of the work and you just review the last 20%. Reviewing broken code isn't like 20%, it's just rewriting code at that point. And if you've used a lot of AI, you've probably wasted tons of time chasing down AI-generated nonsense before realizing you probably could have written this cleaner in about 15 minutes. Sometimes this is the case. Sometimes the fix is actually more AI. I've certainly seen that be the case. But sometimes if you've gone so far out of the boundaries of good code and you've used AI a little too much, you really have to rein it back in and begin writing a lot more code for yourself. Speaking of that, mission critical code is basically off limits. If you have pipelines that ingest millions of rows of data and need to scale and they can't afford inaccuracy, AI can't be trusted to write that. And if it does write that, it needs a lot of human review. The tests are often weak. This is a known pain point from developers. There's actually a guy on the Internet of Bugs, a great YouTube channel you should check out, that addresses this. He's seen and also confirms a lot of what I've experienced as a developer, which is that the tests are kind of lame. They're just pretty, pretty whack, right? The tests are terrible. I've seen tests where it just mocked or basically faked a function and made it return a hard-coded value so the tests would pass. But that's crazy. Who would ever do that? So if it's actually actively trying to fake tests, that is insane. Can you imagine if a human being did this? If a human being did this, they would be fired, almost certainly, right? And here's the most interesting thing that people seem to just skip over AI companies are hiring developers at outrageous salaries. If OpenAI and Anthropic are fighting over the talent and still hiring front-end developers, full stack developers, back-end developers, then what makes you think that your no-name company somehow has the secret sauce? Now here's what the actual research says. Cornell and an organization called Meter M-E-T-R, if I'm pronouncing that correctly, ran a study on experienced open source software developers with AI tools, and they observed that those developers got 19% slower, not faster, slower. The devs actually still thought AI sped them up. This is how strong the hype is. So on one side, you have the AWS CEO saying firing juniors is stupid. And then on the other side, you have researchers saying AI is slowing down senior level developers. And yet, you have a whole internet full of geniuses who, despite literal CEOs of cloud giants and researchers studying top open source developers, will tell you that I've got it wrong, that you've got it wrong, that they've got it figured out how to use AI the right way. Not AWS, not Cornell, no one, just them. Anything is possible if you just believe or if you lie. So recently, I was actually building a data pipeline to ingest hundreds of thousands of TikTok data points. I asked AI to help out with some test cases. At first, the test cases looked okay. They were passing. Everything was green. I'm like, cool, this is pretty good. We had actual data flowing, but the actual data flowing through was quite obviously broken. So I dug deeper. And what did I find? It looked like the AI hard-coded the return values. In other words, the tests were designed to always pass, no matter what. If I had not checked this, we would have shipped something out that wasted tons of API credits, dumped millions of rows of junk data into our system, and left me with a job over the weekend to clean up. Now, obviously, my job is not just rubber stamping AI code or just taking whatever it spits at me, but ask yourself this would any junior developer do something this nefarious? What if you saw your human colleague make this same quote unquote mistake? This isn't like an oops. No, this is actually dangerous stuff. And this is why AI can't be trusted blindly, especially on critical systems. Now, this is not to say it has no place at all. I'm generating the majority of my code with AI. I'm taking a lot more time to plan and think through how it should work, though, rather than just saying you figure it out. I'm giving it very, very clear instructions for the most part, unless I'm doing something like prototyping, where AI, in my opinion, shines the most. When you're doing a prototype, some sort of greenfield work, AI is completely amazing. Chef's kiss for that kind of stuff. I love it. I can build a whole front end, a whole web app with a back end in like minutes or whatever. And then I can tear it apart and do it all over again without worrying about anything. But at some point, when the code becomes just a little too complex, you have to really ease up off the gas. And now I'm using AI to write considerably less code in that same code base than I was when we first started. And if you truly think AI is a one-to-one replacement for developers, don't argue with me online. Just test it yourself. I want you to do this little exercise here. Clone a large open source project and then use AI to try to make a meaningful change. Even better, try to implement a feature or fix a bug and prepare it for a pull request. But please, for the love of Bob, don't actually submit this pull request because you're probably gonna drown these open source developers who are already holding up the entire internet on their backs for free with your junk PR. So don't really make a pull request. But if AI was truly a total replacement, these projects would already be fully automated, in my opinion. And they're not. And if you try to do this, I doubt you're gonna get very far without running to the same problems that the rest of us do. I think this little experiment will give you a very clear picture of AI's actual capabilities, far more than any headline, hype cycle, or even a podcast like this could possibly offer. So where does this exactly leave us here? I feel like we're living in two different realities. We have the hype reality, the idea that AI is replacing jobs, that productivity has doubled, and that junior developers are completely toast because of AI. And then we have our lived experience. AI is helpful and it's messy. Productivity is marginal or maybe worse. And firing juniors in general is dumb. It's like shooting yourself in the foot. Who do you think replaces the seniors? I'm 41. I don't think I'll be writing code in another five years, in another 10 years, almost certainly won't be writing any code. And many people much younger than me are gonna go into other management positions as well. So where do we find the new batch of people? Do we just have AI do it all despite all the madness and mistakes that it could make? Who's gonna look over the AI and know what's good versus bad? Do we just not promote any seniors ever? I don't believe this is a path to do anything at all. This is short-sighted at the very best. Now, at the same time, I'm really, really having a ton of fun working with AI. And that's why this week I'm in Reno, Nevada for one week making the new curriculum for parsity so we can introduce AI, practical AI stuff, mostly with retrieval augmented generation, which I've spoken about at length on this podcast, because I do see it as the thing that web developers should know. It's not a sexy topic, it's not really a fun topic, but it's one of those topics and one of those skills which is gonna get you paid and it's gonna be a lot of fun to learn. So I'm super excited. I'm hold up here, building out the curriculum, and I cannot wait to release it. If you want early access or if you want to actually purchase the material and you're not in parsity, reach out to me. I'm still deciding how that will actually look. So before you jump into another Twitter fight about AI or go on LinkedIn and tell me I'm an idiot or something like that, build something non-trivial. Build on a team of more than three people. Try to update a legacy code base that's been around for more than one year with AI. Push it to the limit until you see it break. And then you tell me what you think, right? Am I full of it? Or do you actually now see the difference between the hype and the reality? Are they one and the same? Are they not? Ultimately, you need to make your own decisions on this thing. But I think that the majority of us who have been in this career for a little bit of time and have some experience are seeing that it's not all that it's cracked up to be. No shade on AI, of course. I mean, it's an amazing tool. We're very lucky to live in this period of time, or maybe we're very unlucky to live in this period of time. I don't know. Anyway, give it a shot, tell me what you think, and I'll see you around next week. That'll do it for today's episode of the Develop Yourself podcast. If you're serious about switching careers and becoming a software developer and building complex software and want to work directly with me and my team, go to parsity.io. And if you want more information, feel free to schedule a chat by just clicking the link in the show notes. See you next week.

Podcasts we love

Check out these other fine podcasts recommended by us, not an algorithm.