The Messed up Part About AI Nobody's Talking About

The messed up part about AI that nobody's talking about is how it slows down senior engineers. While it may be great at "Hello world" and simple algorithms, many times, senior engineers are way past this, and don't really need help building boiler plate code because it just doesn't scale.

I find myself more often than not, fighting with AI in some cases rather than using it to write stuff for me that I need done quickly. From custom algorithms or complex situations where scale, multiple contexts or hidden details about how computers and coding works that you don't gain from the RactJS tutorials, even design patterns that are gained from learning other languages cannot be captured and implemented by AI just yet (learn how C and C++ works and then tell me you know how to code properly in JavaScript) to its inability to write in Angular vs ReactJS or even use a different language other than Python for a product solution. We all know JavaScript and Python are not the most efficient languages out there, so why limit your solution to something as inefficient as these? They are great tutorials and gateways to learning how computers, systems and coding works, but let's be real: C is where you really learn how software works, performs and optimizes and scales.

As such, I find myself spending more time documenting my work, expanding out my timelines and still fighting with the inline suggestions to get the product out the door. Don't get me wrong: It's great that I find myself documenting even the most obscrure algorithms in my code. This helps future me understand the "why" behind a way something was done. Let's be real tho: I'm just better off leaving the auto-completions off or limiting the context to just try to complete the line, not the whole paragraph of code. This goes for llamacoder and copilot. I perfectly documented the solution, explained each of the edge cases, still had to go through the data myself because it doesn't know how to categorize anything and it still failed to perfectly build me a solution. As such, I find myself going into the depths of the product to find out where it went wrong only to discover it still writes code that even a 12 year old could do better (and trust me, I've seen some pretty sharp 12 year olds who made me feel dumb and insufficient).

It was cute at first, but over time, the cracks are starting to show and I am still met with limitations. My opinion remains the same: ChatGPT, Copilot, Gemini, all these coding models still can't code.

So, go ahead, Mr CEO, claim that all code will be written by bots and tell people they can't code better than 60% of software engineers out there. Meanwhile, the top 40% of senior engineers out there already know that 90% of people just can't code. That's not arrogance, that's just a fact. Play helpdesk for a year and you will see just how inept people are too.

In a year or two, when you realize the edge cases you didn't consider for your business weren't handled and vibe coding just doesn't work to handle those edge cases and AI invariably fails to auto-complete for those edge cases, senior engineers will double their prices and have to be rehired back into the workforce to clean up the mess you refused to acknowledge years ago when we told you so.