We Need To Talk About The Vibe Coding Pandemic...

追加: 2026-05-09

[00:00:00]In February 2025, a guy tweeted two words, vibe coding, and accidentally broke software engineering forever. That guy was Andrej Karpathy, co-founder of OpenAI. By the end of the year, 92% of US developers were using AI coding tools daily. 41% of all code on Earth was AI generated, and Collins Dictionary made it their word of the year. 12 months, that's all it took to take vibe coding from a casual tweet to a full-blown pandemic. The first people to feel it were junior devs, because of course. US junior developer job openings are down over 67% in the UK. Entry-level tech roles dropped 46% in 1 year. Salesforce straight-up stopped hiring software engineers altogether. I'm sure that won't have any negative impact on the company when the AI bubble bursts. Any day now. Any day. But here's the part that's actually scary. The juniors who are getting in are falling into what I'd call the new tutorial hell. You remember old tutorial hell. You'd spend 40 hours watching YouTube courses, feel like you understood everything, open VS Code, and immediately understand nothing. Vibe coding hell looks different on the surface. You're actually building stuff.

[00:01:13]You're shipping things. It looks like progress. But you hit a bug, you paste it into Claude, you get a fix, you move on. You never learned why it broke.

[00:01:21][music] You just vibed past it. And the next time you hit that same bug, you paste it in again. And here's what makes it worse. AI doesn't close the skill gap. It widens it. Employment for developers age 35 to 49 is up 9% for 22 to 25-year-olds, it's down 20% from its 2022 peak. Because a senior dev uses AI to knock out boilerplate fast, and then handles the architecture themselves.

[00:01:46]They read what the AI wrote the way they'd read a junior's pull request, with suspicion. A junior dev uses it for everything. And when something breaks in production, they have no mental model to debug from. They just asked the AI again. Now, here's where it gets interesting because there's another side to this. 63% of people vibe coding right now are not developers at all. A warehouse owner in West Virginia built custom shipping automation in a single day, and this isn't an edge case. The definition of developer is slowly starting to change. You don't need a CS degree, you don't need to know what a pointer is, you just need to know what you want to build. That's literally it, which sounds amazing until you ask the obvious follow-up question. If anyone can build anything, and most of them have no idea how the underlying code works, who's actually checking what they're shipping? Because here's the thing, developer trust in AI [music] tools dropped from 77% in 2023 to 60% today, and usage keeps climbing anyway.

[00:02:45]The industry knows it has a problem and genuinely cannot stop. And if you want a timestamp for when the cultural stigma around this officially died, it's January 2026, when Linus Torvalds, the man who personally wrote the Linux kernel by hand in 1991, used Google's AI tool to vibe code a Python visualizer and documented it openly in the project readme. If Linus is doing it, the conversation is over. So, let's talk about what this looks like when it goes wrong. Lovable, one of the biggest vibe coding platforms out there, had 170 out of 1,645 of its generated apps shipping with vulnerabilities that exposed personal user data to the More than one in 10 apps. The platform did what it was built to do. The code just didn't protect anyone using it, and that's not an isolated incident. AI-assisted code commits leak secrets at twice the rate of human-written code. In 2025, GitGuardian found 28.65 million hard-coded secrets sitting on public GitHub, up 34% in a single year.

[00:03:49]AI service API key specifically were up 81% we have an entire wave of apps being shipped by people who don't fully understand what they built running on infrastructure that wasn't designed to be this exposed. The pandemic isn't just in the code, it's in the attack surface.

[00:04:05]And like I said earlier, this vibe coding pandemic it's even scarier when you realize most cybersecurity professionals don't have hands-on training for these AI specific threats.

[00:04:16]That's why I've partnered with TryHackMe, the world's largest hands-on cybersecurity training platform to help combat us. They've launched a new AI security learning path with 26 hands-on labs across five modules where you practice on real AI systems right in your browser. No shitty installs, no expensive hardware, no setting up your own infrastructure. Just open your laptop and start learning. You also don't need prior AI knowledge. It takes you from fundamentals in securing AI systems to prompt security, AI supply chain attacks, and data poisoning. By the end you've actually broken and defended AI systems yourself, not just watch someone else do it. It's fully self-paced and hands-on, so you're building real skills. Start for free at and use code code head AI for 25% off the annual pro plan plus an extra 15% off certifications. Thanks to TryHackMe for sponsoring this video.