Agentic coding is great. I have been leveraging this technology daily for past six months. I’m not here to discuss its precision or the productivity gains it offers. As the saying goes in the AI world: this is the worst it’ll ever be.
I am concerned about something completely different. The dopamine effect of triggering the LLM through a click of button, whether it’s for writing, refactoring, or debugging code. It feels like a slot machine, except here your odds of winning (getting a good solution) are nine out of ten. And soon, that’ll become ten out of ten. This whole vibe coding has gamified some part of software development.
When you spend 10 hours debugging a code or refactoring a code, you learn a lot of things in the process, like design principles, code quality concepts, better awareness of common pitfalls. This leaves you happy and satisfied in the end when you see the result of your effort. But with vibe coding it’s all about instant gratification. Triggering LLM once or twice (if the result is not right) takes the same effort. Devs are using prompt to get the solution and another prompt to explain the solution and then next prompt to write the test case for that solution.
This feels like magic wand, the developers are now 10x productive, even product managers can now build a quick POCs to validate their ideas, most of the infrastructure work can now be automated under a single person which earlier required a complete team. But there is a tradeoff. We’re sacrificing the constant learning and growth that naturally came from the struggles, struggles now replaced by these tools..
I align with the belief that 90% of routine development work should be automated, and that human creativity should be focused on the remaining 10% areas where deep insight is needed and where LLMs still fall short (because of lack of training data). But the reality is that many used to build their capabilities by grappling with that 90%. It’s through that struggle that developers earned the depth needed to contribute meaningfully to the 10%. With that learning opportunity fading away, we need to find new ways to train ourselves and stay relevant for the remaining 10% of work that still requires human developers.
Top comments (0)