Jan 1, 2026

AI driven development has different priorities

Late night rant about how AI does not give a f*ck about you
rant

Normal day of using AI to get done with my Jira tickets as soon as possible, and I noticed something — like those revelations you get while taking a shit.


I was working on a permission-based UI; basically, different options for different roles. Backend sends a JSON with all the permissions and boolean flags. Simple stuff, right?

So, I thought of making a global context plus a utility function coupled with interfaces, and that should be enough work as per my salary. Explained it to Cursor and went back to watching cute girls on Insta.


It came up with something that worked, but what should have been a simple Array.filter was far more complicated than necessary. And that led me to a thought.
What if AI starts writing code that works, yet is hard to understand, review, and maintain?

That would defeat the whole purpose of high-level languages — which exist to make sense to dumb humans, even at the cost of speed.

I have seen this issue with code generated by tools like Lovable, where the output works but god knows how.

Since LLMs are language models, well-documented code with intuitive variable names should be enough. but does that exist in the codebases?

And no, the solution is not to litter the codebase with comments just to assist the AI, at the cost of human readability.

Priorities

  • Token efficiency over human readability: these systems are trained to minimize tokens because anything else costs money. The question is whether we are actually trying to strike a balance, or silently giving up readability.

  • Gibberish is harder to verify: unreadable code is harder to reason about, which means more time spent testing and debugging.


If we start using AI to verify AI-written code, the circle of AI slop will be complete. And when production is down and you ask why, the AI will start with, “Great question!”
Guardrails like MCPs and Cursor rules have helped me, but the real question is longevity. Six months from now, when an urgent bug shows up, will there be even one developer who knows the system end to end, remembers the file name, and can fix it fast?

I found some resources matching this:


Happy vibe coding!!!