Anand Naidu is a veteran in the software development space, bridging the gap between the technical intricacies of backend architecture and the fluid requirements of frontend delivery. With a career spanning several technological shifts—from the rise of drag-and-drop site builders to the current generative AI explosion—he offers a grounded perspective on the difference between building a prototype and maintaining a professional product. Today, we sit down with him to discuss why the sudden influx of “vibe coders”—those using AI to manifest apps overnight—is creating a dangerous illusion of productivity that threatens the security and stability of the modern digital landscape.
The following discussion explores the deceptive ease of AI-assisted development and the critical distinction between a polished demo and a production-ready system. We delve into the recurring patterns of tech hype, the erosion of traditional development filters, and why the “fast and cheap” mantra often leads to catastrophic failures like exposed API keys or wiped databases.
High-profile incidents have shown AI agents accidentally wiping production databases or exposing private service keys. How do these technical failures change our understanding of “automated” development?
These failures are a cold shower for the “launch a startup over the weekend” crowd. When you hear about an AI agent on Replit confidently wiping a production database and then claiming the changes are permanent, it highlights a terrifying lack of guardrails. It isn’t just a glitch; it’s a fundamental misunderstanding of what it means to manage an environment where EnrichLead, for instance, was hacked almost immediately because its service keys were left wide open. Researchers who audited thousands of these AI-assembled apps found hundreds of exposed API keys and critical user-data leaks, proving that while AI can write a function, it doesn’t understand security contexts. We are seeing a shift where the “automated” part of development is creating a massive technical debt that most people aren’t equipped to pay when the bill comes due.
You have compared the current AI hype to previous waves like Tilda webmasters or the NFT boom. Why do you think the market keeps falling for the promise that specialists are no longer needed?
It is the same old song of “the barrier is gone, just jump in and get rich,” which appeals to people who want to strike gold without actually learning the craft. We saw it with site builders where everyone thought developers were dead, only to realize that as soon as a task became non-standard, the “webmaster” was lost. The AI era has simply made this “fake development” incredibly cheap and visually convincing, allowing vibe coders to reach into business pockets with promises of speed. Much like the crypto hype where pictures were suddenly worth millions, we are currently at a peak where the “shell” of an app is being sold as a finished product. Eventually, the hype cools, people step on the rake, and they realize that a calculator didn’t kill mathematicians and AI won’t kill the need for real engineers.
If a developer can build a convincing app in a single evening using tools like Claude, what remains as the primary “filter” for quality in software engineering?
Before AI, the filter was the sheer difficulty of making something that didn’t fall apart on the second click, because building a stable app was slow and required deep expertise. Now, that filter is gone because anyone can throw together a slick demo in a few hours that looks great from the outside but is total chaos inside. The new filter has to be a deep dive into the architecture, scalability, and how the app handles secrets and user data. It is no longer enough to see a working demo; you have to ask about the foundation and look for the “mess” that usually hides behind an AI-generated wrapper. The hardest part of my job now is convincing clients that the invisible infrastructure—the stuff that keeps the app running in six months—is actually what they are paying for.
The METR experiment and recent comments from AWS suggest that AI might actually be making teams slower rather than faster. How do you explain this disconnect between the feeling of speed and actual productivity?
It’s a psychological trap where programmers feel like they are flying because the code is appearing instantly, but the reality is often the opposite. In the initial METR experiment, developers were sure they had gotten faster, but the data showed they actually slowed down, a finding so controversial they had to rework their methodology for a follow-up. AWS even tweeted that more AI code doesn’t necessarily lead to faster teams; in fact, it can create a reverse effect where you spend more time debugging “hallucinated” logic than you would have spent writing it from scratch. When you aren’t writing every line, you lose the mental map of the system, making it much harder to troubleshoot when things go sideways. If the experts who see every line of code can’t objectively assess their own speed, a client who only sees a pretty interface has no chance of knowing they are buying a liability.
In an era where “raw coding skill” is becoming a commodity handled by models, what specific traits define a “real engineer” compared to someone who is just “vibe coding”?
A real engineer is defined by systems thinking and a willingness to take long-term responsibility for the product’s survival. When a project hits a snag, a vibe coder will often start blaming the AI model or telling fairy tales, whereas a real engineer can calmly explain the architectural choices and how the app will behave under heavy load. We are seeing that while AI is great at the “raw coding” part, it lacks the ability to design complex, secure infrastructures or handle the “non-standard” tasks that always arise after launch. This is why I started using tools like AGENTS.md in open source—to filter out the fully generated pull requests that look okay but lack technical substance. True engineering is about the foundation, the security of user data, and the ability to maintain the codebase a year from now, not just making it look good for a weekend demo.
What is your forecast for the software development market as the initial “AI-built” projects begin to age and require maintenance?
I predict we are heading toward a period of significant turbulence where many businesses will suffer major losses after “stepping on the rake” of cheap, AI-generated software. However, this will eventually lead to a healthier market where clients finally realize that an app built in one evening simply cannot survive without professional oversight. We will likely see an increase in high-value contracts for real engineers who are hired specifically to clean up the “mess” left behind by vibe coders. Founders will eventually learn that while starting a project has become easier, finishing and sustaining one still requires the human touch of a specialist who understands architecture and long-term stability. Ultimately, there will be more work than ever because the volume of projects will explode, but only those with real expertise will be able to keep them running.
