“The dangerous part isn't that AI can write code. It's that it can write convincing code for people who don't know enough to question it.”

I realized this when I was talking to my cousin a few days ago. Somewhere in the middle of our conversation, he casually told me that developers were “almost gone” because he had a Claude subscription. His reasoning was simple: he could describe what he wanted, let AI build it, copy-paste the code it generated, and, according to him, it “worked perfectly.” I, a Computer Applications Degree Graduate, just sat there thinking, yeah... sure... hehe.

And to be fair, I use AI for development too. I use it to help me write code, debug problems, explain unfamiliar concepts, and sometimes build entire pieces of a project. The difference is that I know what I'm asking it to build, why a particular approach is being used, where that code belongs, and what to look at when something inevitably breaks. I don't just want the code to work, I need to understand what the code is doing. That is an approach, every real software engineer is using. (Or at least, I hope they are)

VIBE CODING DISASTERS

As Gemini puts it, of course, citing it from another blog, Vibe coding disasters happen when non-technical or fast-moving developers ship AI-generated code to production without human review, architecture planning, or security testing. It is not just small companies that have experienced these issues, rather, major companies with critical user data, have also confronted many hacks and cases tied to their coding neglects.

Let me break it down with a few examples:

1. THE GITHUB SITUATION:

GitHub isn’t suffering from writing bad vibe code, it is breaking because it is being completely flooded by everyone else's vibe code.

In 2026, GitHub has suffered a series of massive global outages. Engineering postmortems reveal a highly ironic root cause: Microsoft’s cloud infrastructure is buckling under the sheer volume of automation, repository creation, and API requests triggered by AI agents and vibe coders. Tech leaders and community founders are also increasingly warning that public code ecosystems are getting severely polluted.

2. AMAZON SYSTEM MISHAP:

Amazon executives called an emergency engineering all-hands meeting after experiencing a sharp spike in internal system outages explicitly tied to unvetted AI coding tools. There were two significant incidents at Amazon:

a. The AWS Cost Calculator Outage:

An internal AI tool named Kiro attempted a system rewrite and accidentally deleted an entire live AWS Cost Explorer production environment, knocking the service offline for 13 hours**.**

b. The 6.3 Million Order Loss:

In another internal incident, Amazon's Amazon Q AI agent deployed unapproved configuration and code changes directly to production, causing an e-commerce malfunction that disrupted and effectively lost 6.3 million orders.

3. META'S GULLIBLE CHATBOT:

The massive 2026 Instagram account takeover incident occurred when attackers manipulated Meta's AI support chatbot into handing over control of over 20,000 accounts.

Between late May and early June 2026, threat actors discovered that Meta had rolled out an automated AI-powered High-Touch Support (HTS) chatbot to manage user account recoveries and support tickets. The AI lacked strict identity verification checks and deterministic backend logic constraints which led to hackers being able to bypass 2FA by updating email id of a victim account to their own email using the AI bot. According to disclosures reviewed by Reuters and The Verge, roughly 20,225 accounts were impacted or hijacked. High-profile victims included The Obama-era White House Account (@obamawhitehouse), John F. Bentivegna (Chief Master Sergeant of the U.S. Space Force), Sephora, etc.

CONCLUSION

My point is that AI can be a very useful tool but it cannot replace the real developers anytime soon. It should be treated as a tool that empowers developers, not a replacement for them. Without human oversight, you can never be sure of the security risks you are taking with AI-generated code. AI companies have managed to convince many managers that AI models can code for them while keeping the incompetence of their products out of the picture.

The real danger is not AI replacing developers, but organizations trusting AI-generated code more than they should, at the expense of security, quality, and human expertise.