AI is CRASHING and Software Engineers Got Their Opening. transcript

You probably heard a version of this and I know you definitely heard the other side of it. But what a lot of people don't have is data and anecdotal data at that. But I have that because I've seen software engineers within my own mentorship land interviews and land actual programmer jobs multiple dozens of times in the last year alone. So when I say that AI is not only going to crash, but also software engineer openings are being created.

I am speaking from what I've seen with my own two eyes. So how is AI crashing? Well, it began with a company in Louisiana called Veneer Investor. You've never heard of it. There's no product. There's no app. As far as I can tell, there's barely anybody who works there. It has borrowed about $27 billion. $27 billion from PIMCO, from BlackRock, from people who do not lend that kind of money to companies with cute names unless they're extremely confident about something.

And Meta owns 20% of it. Now, that company is building an enormous data center, 4 million square feet, enough power for a small city. And when it's finished, it will rent that building to Meta. So Meta gets the data center. Meta does not get the debt. None of that is illegal. None of it is hidden. It was in a press release. But if you open Meta's balance sheet tomorrow looking for $30 billion of risk, you would not find it because it isn't there.

It's in Louisiana in a company you've never heard of that Meta owns a fifth of. So I'm going to come back to that later because it is related to you. Because you're probably thinking that this market is cooked and that AI has completely replaced the job of software engineering. So let me come back to that, but I can't explain why none of this matters until I tell you about a guy in my program who could absolutely build anything and couldn't explain a single line of it.

Because the reason he froze in an interview last spring and the reason that 27 billion is structured that way are the same reason. So, let me start with him. I'm not going to use his name and you'll understand why in a minute. If you are watching this video, you know who you are. Anyways, he joined a year ago and he was fast, genuinely fast. The first week he showed me a full application login payment to dashboard and he put it together in about 4 days and I remember thinking, "Okay, this is the hype behind the tool everybody's been telling me about." He wasn't writing much of it.

He was describing what he wanted, reviewing what came back, and shipping it and it worked. So, we got on a call and I did what I do, which is ask, "Why?" I asked him, "Why did you handle logins that way?" And he said this, and this is the exact phrase, "I think that's just how it did it." Not, "That's how I did it." That's how it did it and I let it go. The things worked. He was moving faster than anyone in his group.

And I didn't want to be the guy who kills the momentum over something philosophical. That was a mistake. And it's my fault as a mentor because about 3 weeks later he hit a bug. Small one, something in the payment flow firing twice under a specific condition, and he couldn't fix it. He genuinely couldn't even with the help of AI. It just kept getting messier and messier. And yes, we did fix it eventually together, but it took a few sessions.

Let's just say that. Because to fix a thing you have to know what it's doing and he never known what it was doing. He only ever knew that it worked. So, he did the thing you do when you can't fix something you don't understand. He rebuilt it, and the new version had different bugs. And I told him to go back to class from the beginning. He lost 2 months in that loop. And just like the majority of our guys, within those 2 months, he did land an interview.

We were genuinely excited. Many was already asking for a testimony, and the mentors were hyping him up like it was already a done deal. He got an interview with a really good company that would for sure pay him a bag. And they did something more companies are doing now. Pulled up his own project, shared their screen, and asked him to walk them through one decision he'd made in it. But then he froze. Not because we didn't go over it or anything, but because it wasn't his answer to give.

He gave the answer that we came up with collectively. But he just regurgitated the answer he had memorized during our session rather than studying it. Eventually, the rejection came a few days later with one line in it about wanting more depth. And he sent me a screenshot of that email at 1:00 in the morning with no message attached. Just the screenshot. If you've been there, you know exactly what that message meant.

I'll come back to him, too, because the story doesn't end there, and the ending is why I'm making this video. Let me go back to Louisiana. I'm not an economist, and I'm not going to pretend to be. I'm just a software engineer with a family. So, I follow this stuff because it's literally my job nowadays to help my mentees succeed in this market. So, let me finish this story. Meta is not a company that struggles to find money.

So, why borrow it through somebody else at all? Because $30 of debt on your own balance sheet looks bad. Analysts see it. Your stock takes it personally. So, the debt lives in Louisiana. And the building gets rented back. Now, what goes inside the building? Chips. Nvidia chips, because Nvidia sells most of them. So, watch the money go round. First, Wall Street lends to the company in Louisiana. Second, Louisiana builds the building.

Third, Meta rents the building and fills it with Nvidia hardware. And finally, that purchase shows up on Nvidia's books as revenue. Nvidia stock goes up, and Nvidia stock going up is most of the reason you believe this boom is real. So, you think about that for a second. The borrowed money buys the chips, and the chips being bought is the proof everybody points to that the borrowing was smart. Every step of it is legal.

Take any one on its own, and it makes complete sense. And that's one building. The Wall Street Journal added this kind of thing up across Big Tech and got to around $3 trillion. Now, let me tell you what bothers me as a software developer myself. A building lasts 30 years. The chips inside it don't. Not only are the old chips burning up and less effective, the new chips are able to do more work on less electricity. So, the old chips get outdated like every 18 months.

But to be generous and to be realistic, let's say about 3 years is how long a chip stays economically worth running. However, on the books, the cost of these chips that last 3 years gets spread over 5 or 6 years. Therefore, spread a 3-year cost across 6 years, and your profit looks better than they actually are. And it's paid for with debt shaped like a building. Now, I say all this to tell you that this isn't fraud.

It's timing. What is happening is enormous money is being spent now for money that they expect to make later. Amazon is famous for losing money on AWS for years, and eventually, it turned into one of the most profitable businesses on Earth. So, if AI starts earning what these companies are basically betting it'll earn, and it does it soon enough, the loans get paid off and none of this was a bubble. So, that's actually what's being wagered here.

AI is impressive, yes, but it's all about whether it makes real money fast enough to cover payments that are already due. I tell you all this because it is this question, will AI make the money fast enough that decides whether people keep their jobs. Now, I need to tell you my own version of this because it's how I know what a story like that does to people. 2024, ordinary Tuesday, a calendar invite shows up in the morning with no agenda on it.

And by the time I joined that call, I already knew. My whole team was there. And if you know my story, we'd spent months building a social media app for a coffee machine company that distributes coffee machines in Korea along with many other small products as well for that specific company. It was the biggest app that I had been a tech lead for, for sure, with 250,000 people in queue for it. Well, long story short, the company let us go and I couldn't believe it.

If you scroll way back to the beginning of my channel, that's the video I filmed on my phone right after this meeting had happened. Part of it was, of course, I always wanted to start my own YouTube channel and always put it off because I was always going to work at 7:00 a.m. and coming back home at 9:00 p.m. Now, I had no excuse not to record. But, the other part of me was just straight-up scared for my life, wondering what was next for me, so I just hit record to document my emotions.

Within about a week, the explanation was everywhere. Internal comms, LinkedIn, the news, all of it saying the same thing and talking about cost efficiency and AI being the reason for the layoffs. And I was in that codebase. I know exactly what our AI usage looked like in 2024 because I used it every day. It was useful. It sped me up. On a good day, it was like having a fast junior who never got tired and occasionally lied to me.

But, it was not doing the same work of a whole team. What actually happened is we've been building for a year or so and the parent company's real business was selling coffee machines. At some point, somebody upstairs looked at what we cost and asked a completely reasonable question. What is a coffee machine company doing running a social network? That's the real reason. But, AI made us more efficient is a wonderful sentence to say to a board.

Now, put that next to Louisiana. Investors handed over billions of dollars not because of what AI can do right now, but because of what they think it'll be worth in a few years. Here, the debt is just parked in a separate company in Louisiana. Now, my old company sunk a year into a social media app that sells coffee machines and it didn't work. So, instead of saying that it didn't work, they said AI made them more efficient and cut the whole team.

In both cases, one larger scale than the other, something big and uncomfortable gets relocated somewhere it won't get inspected. And it's nothing illegal, either. So, these two cases work together to increase the stock of AI because investors put billions in on the promise that AI will replace expensive workers and they want to see it happening. Meanwhile, ordinary businesses are cutting staff for ordinary reasons. They over hired, a project failed, a bet didn't pay off.

Nobody wants to say that out loud. It makes leadership look bad, so they say AI did it. Then, the stock of AI goes up. Which is exactly the evidence the investors were waiting for. So, more money flows in. The promise looks more credible. And the next company that needs to explain a layoff has an even more believable excuse ready to use. This cycle is what a CS student hears, or a beginner hears, and they feel like there is no point.

So, let me tell you what I think actually happens next. If this money story falls apart, the next year or so will get worse, not better. Companies that only existed because AI was fashionable will close. Jobs with AI in the title stop paying a premium, and hiring slows down. Because when everyone's scared, nobody makes decisions. But look at what's standing afterwards. It's parallel to the dot-com bust. 25 years ago, phone companies were certain the internet was about to explode.

So, they buried fiber optic cable, the stuff that carries internet traffic all across America. Enormous amounts of it, and the traffic didn't arrive fast enough. The companies ran out of money and went bust very publicly. But here's the thing, the cable was already in the ground. Nobody dug it back up. It just sat there, nearly worthless, for years. Then, a few years later, someone looked at all that unused cable, realized it was now dirt cheap to move video around, and built a website where anyone could upload video for free.

That's YouTube. The thing you're watching this on only exists because somebody else lost a fortune first. That's what a bubble does. The world massively overpays for infrastructure, goes broke, and hands the infrastructure to whoever shows up next. So, when I say this pops, the valuation pop, the story pops, the data centers don't evaporate. The models don't get deleted. The chips keep working. They just get cheap, which is the best thing that can possibly happen to somebody trying to get into this industry with no money and no connections.

And I want you to think about who's actually holding this industry together. It's not the famous people. It's the senior developers who know why the old system is weird, who can look at an outage and know where to start. They are already 10 years into the game, but you can't hire more of them. We need to start growing them now because a senior engineer is just a junior who broke things over and over and over again and had to sit there while it got fixed.

That takes a few years and there's no shortcut. And the problem gets worse because companies have barely hired any juniors in the last few years. We don't necessarily feel it yet, but we will feel it in about 8 years when there is no one with 8 years of experience left. It happened in 2008 and by 2011, companies couldn't find mid-level engineers anywhere. Therefore, pay shot up for the few people who stayed through the bad years.

And I do believe this is the reason why stats show that job openings for software engineers are on the rise. So, put it together. The tools will get cheaper and at the same time the industry will start asking for developers on the come up. That's the window for those trying to become software engineers today and it's opening more and more every month. But these engineers are different than the engineers that came before us.

You have to be the right kind of engineer when this window opens. Here's what actually changed in the last 18 months. Producing code stopped being the valuable part. For 40 years, the scarce thing was somebody who could turn an idea into working code. That was the job. That's what interviews tested and that skill got commoditized, not eliminated, commoditized. It got cheap. What did not cheap is judgment. Ironically, it's the skill that senior developers need the most of.

And let me make that concrete because it sounds abstract until something breaks. I reviewed some code a while back, honestly better written than most of mine, and it did exactly what it was supposed to do. But say two people hit it at the same moment. Both of them look at the number, both see the same thing, both add one, both save. Two people just did something and the number only went up by one, and nothing breaks.

There's no error, nothing in the logs, nobody gets a warning. The number's just wrong sometimes, and only when the site's busy. Now, here's the part I want you to get. You cannot find that by checking whether the code works. It does work. You find it because you already know what pattern exists and you went looking for it. And you only know it exists because at some point you shipped one, and you spent a very bad week working out why the numbers didn't add up.

That's the job now. In my case, it was knowing which of the four approaches survives contact with 250,000 users. Reading code you didn't write and working out what it actually does versus what everyone assumed. Looking at output from a model and knowing, not guessing, knowing that it's subtly wrong and being able to say why. Understanding what the business is actually trying to do so you can tell them when they've asked for the wrong thing.

Every one of those is a reviewing skill. That's the new job. You're not the person producing the work anymore, you're the person responsible for it, which is a promotion, honestly. It's just a promotion nobody prepared you for, and it's the exact skill set the last 3 years of content told beginners they could skip. Now, going back to my mentee, again, you know who, this is why he got burned at the interviews. He had the production skill at a level that genuinely impressed me, but he had none of the judgment.

And the market has stopped paying for the first without the second. Now, let me tell you what changed in my program and what my mentees are doing right now to be ready for 2027. We put people in code they didn't write. Everybody used to start green. Your own project, your own decisions, empty folder. That's how everyone teaches and it's how almost nobody works. Employment is the opposite. You, in somebody else's mess, on a deadline, staring at a decision made in 2019 by a guy who left and documented nothing.

So, people spend real time in unfamiliar codebases now. Find a bug they didn't create, reproduce it before touching anything, which sounds obvious and almost nobody does. Then, fix it and explain the reasoning like they're talking to a tired senior with 40 tabs open. Do that a handful of times and you're in a small group because the whole rest of the internet is teaching greenfield. AI isn't banned. It's supervised. I get asked constantly whether we let people use it.

Of course, we do. Refusing is career suicide and anyone telling you otherwise is selling nostalgia. The rule is one line. Use it as hard as you want, but never ship anything you couldn't defend in a code review. And we enforce it, which is the part that's different. Now, you might see me in the day in the life videos not using AI. I let my AI workflow mentors handle teaching AI workflows to my students. What I am helping them do, because I have the most experience doing this, is helping these guys think, create solutions, and make judgments on hard decisions on the production grade apps we create together.

On a call, you don't get to show me the app. You explain the decisions. Why this and not that? What breaks at 10 times the traffic? What happens if this request arrives twice? Then, after we go through this, we actually encourage AI because that's the new way now. We teach that too. But learn AI properly after you're able to do what I explained above. And next, everything gets built for real person, not a portfolio piece, somebody whose phone number you have.

Chris built a menu app for an actual food truck. Real food truck, real owner. What got him hired wasn't the app. It was that one night the specials page wouldn't update with a queue out front and the owner rang him at 9:00 at night. And he had to fix it while a man was losing money at him. Months later an interviewer asked about the project and he didn't list features. He told them about the phone call. And that's the difference between a project and an experience.

A project is something you finished. An experience is something that went wrong and you're the one who had to fix it. Interviewers can tell in about 40 seconds. AI has never had an angry food truck owner on the phone at 9:00 at night. And you aim before you build, which is the ordering almost everybody gets backwards. Gabe built a spending tracker, the most generic project on Earth. And that's exactly why I'm telling you.

The idea was generic, what he did with it wasn't. He picked the company he wanted first, went and studied how American Express did this in their own product. So when he sat in an interview at American Express, he isn't asking them to imagine him doing the job. He'd been doing the job. He'd been doing a version of it for months. Most people build something and then go looking for a job. He picked the job and built toward it.

Then there's Mac who's the other end of it. Three years, around 2,000 applications, and heard back nothing. 2,000. It was never effort. He had more effort than almost anyone I've worked with. It was the wrong plan executed heroically. What changed was targeting roles where his background was an advantage instead of a gap and treating what he built for local businesses as actual work experience because that's what it was.

He got hired at an AI startup afterward. None of those built an impressive idea. A menu app, a spending tracker, a job board, on paper, beginner projects. What made them work is each one of them came from somewhere real and gave the builder something to say in a room that nobody else in a power could say. So, back to my mentee, he didn't quit. He went back to that first project and spent about 6 weeks going through it properly.

Not rebuilding, reading. Working out what every part actually did and where it would fall over. By the way, to give you a clue, that project was an interview helper. Then he built the next thing with AI at full speed, but he'd stop every so often and make himself say out loud what it had just handed him in front of our mentors. It took about 4 months before he was a different engineer. And the tell wasn't that he'd slowed down.

He was still fast. It's that when I asked him why, he had an answer. And sometimes the answer was, I don't know yet. I need to go and check. That's a senior answer. That's what a senior sounds like. He's employed now. The tools made him fast inside a year. The market wanted the thing the tools don't give you. And the gap between those two facts is where every job is going to be for the next 5 years. He didn't need to be slower.

He needed to know what he was shipping. One more thing and then I'll let you go. I was 30 when I started and nobody else around me thought it was a good idea. Not because they were unkind, because on paper it wasn't. A man in his 30s with responsibilities starting over at the bottom of an industry he knows nothing about. If a friend told me that today, I'd probably ask him some hard questions, too. And the first few months were bad.

They were really bad. 11:00 at night, after everything else was done, with nothing left, and I was slow. I'd spend 3 hours and have nothing at the end of it. Not a working thing, not a broken thing. 3 hours gone. And I nearly stopped around month five. I just didn't open the laptop for about a week. Then I would go at it again for like 2 weeks. Then I would burn out and another week would go by. That's how quitting actually happens.

Nobody announces it. And here's what I only understood later. It was a terrible time to start. Nobody in my life thought it was sensible. And let's talk today. The mentees that were successful last year and this year, when they started, it was also a very terrible time to start according to YouTube videos, the company reports, etc. There's really no good time to start on paper. But the people waiting for a good time are still waiting.

If it takes 8 to 18 months to get genuinely employable, and that's the real number, and not the 12-week thing in the ad, then somebody starting during the ugly part is finishing right at the time things get good. And the person waiting for the market to look friendly starts competing on the same day. Everybody else has the same idea. Nobody opens the window. You go through it. Now, everything I described about the strategies of what I would do to break into the industry is what we do in the mentorship, and it's constantly changing.

And so, we are also constantly changing it, too. Nobody fails because they didn't know they should build real things. They failed because they built the wrong thing, in the wrong order, alone, with nobody who tell them the truth about their code. And speaking of AI, and everybody already knows AI is here to stay. But do you know how actual companies use AI? And it's not a tutorial on how to write code. You need to learn real AI workflows real companies use.

We do that here. If you want that, go apply at leftfield.com. And one last thing. If you are stuck, whether it's fear, whether you're stuck building your app, or can't get interviews, share with me in the comments with I am starting today. Because today will be that day we turn that around. I'll reply back with my advice and my thoughts on trying to help you break through. Anyways, the money's going to do what money does.

Something priced for perfection gets repriced, and there will be a stretch where every headline sounds like the end of this industry. It isn't the end. The tools get cheap. The infrastructure stays. And somewhere around 2028, 2029, this industry looks up and realizes it spent 5 years not making a single new senior. Somebody has to be there when that happens. And that's why the hiring is happening now, as we speak. But they aren't hiring developers from 2017.

They are hiring people who can read code they didn't write, own it when it breaks, and explain every decision in it. That person has to start now, and it might as well be you. So, just remember, if I can do it, you can do it, too. Coding saves [ __ ] lives.