Coding tools are great. I genuinely think they will eventually change how we build software.
But ask around, and people will tell you they have already changed the way build software. Have they?
That depends on your perception. For me, sure, we have seen some improvement. We can write code faster. We can prototype an entire application in a day or two. But does that really improve the software development process?
I don't think so. Not yet.
1. SPEED != ENGINEERING
We are confusing typing speed with actual engineering
Here is why the current state of AI coding isn't the revolution people claim it is and why the problem isn't the AI, but us.
AI coding tools allow us to generate massive amounts of code instantly, but often, thus just adds massive technical debt to the project. The AI is programmed to follow "best practices" from the internet. It was trained on that data, so it mimics it perfectly.
That sounds great in theory, but in practice? Those generic best practices are often completely misaligned with your current code-base or your organization's specific architecture.You end up forcing a generic solution into a custom problem just because the AI read it on a tutorial from three years ago.
2. THE TOOLS AREN'T WEAK. DEVELOPERS ARE.
I am not seeing that AI coding tools are perfect.I'm seeing that developers are weak.
The tools are powerful engines, but we don't know how to drive them.
Missing Context: We can't give better context to the AI. We treat it like a magic wand rather than a junios developer that needs to understand the why behind the code.
Poor Prompting: We don't know how to write better prompts. We ask vague questions and accept vague, mediocre code in return.
The Review Crisis: Worst of all, developers have become lazy. We don't review what the AI has produced. We see that looks right, hit "Accept," and move on. We are building systems we don't actually read or understand
3. THE "GRADIENT" FATIGUE
This laziness is visible to the naked eye. Look at the recent wave of web apps solely generated UIs with gradients everywhere.
Why? Because the AI only knows that specific, trendy design criteria, and developers aren't skilled enough(or patient enough) to prompt for something different. We are filling the web with homogenized designs because we are too lazy to critique the output.
AI coding tools are only as good as the context you provide and the scrutiny you apply to the output. Until developers stop being lazy with context and start actually reviewing the software development process.
We're just getting lost faster.
Top comments (0)