I have seen AI do wonders when it comes to prototyping a new feature or acting as a sounding board for architectural decisions. However, my core takeaway during the last few years is that AI is a powerful amplifier and not yet a complete replacement for developers.

Here is why I believe we need to approach it with a little caution6:

A) AI relies heavily on the engineer steering it

With the rise of "vibe coding", it is easy to prompt an AI into building something that appears to work on the surface, but the underlying architecture can be incredibly fragile.

Without deep baseline knowledge, two major issues tend to happen:

Hidden Vulnerabilities: AI can miss vital constraints like database access controls. For instance, I have seen a junior engineer use AI to build a feature that looked fine initially, but it contained a severe edge-case bug that accidentally wiped records from the database.

The Debugging Loop: When a bug occurs, a developer without a strong mental model of the system often feeds the errors verbatim straight back into the AI. The tool usually provides quick patches that fix the immediate error but break things upstream, compounding the technical debt.

To safely use AI, you at least need someone with a solid understanding of development fundamentals to properly audit and refine what it generates.

B) The business risks of over-reliance

The risks extend beyond the codebase to the actual business. I previously worked at a company turning over £8 million a year with a guided selling tool based on human psychology that decided after VC backing to aggressively pivot to AI. I actually chose to leave the company before the transition took place back when they were making there first set of redundancies, warning them that the shift needed to be handled properly, but those warnings were ignored.

Unfortunately, the resulting drop in product stability alienated their user base. Since then, the company has lost 80% of its workforce and most of its clients, forcing them entirely back to a startup mentality. It has been only 2 years since I left.

Summary is: AI is an excellent tool for a developer with at least a basic understanding of programming, but it can become a liability without proper oversight. If we do not understand how the code works, we lose control of the software we are building.

I believe that your approach to checking the code that is being written is definitely the right thing to do. I am not resentful of AI but am worried that we are blindly trusting AI even though we know it is not perfect and needs to be passed detailed instructions in order to get it right.