Before I even knew what “vibe coding” meant, I had already vibe coded my first app. In fact, anyone who has used AI to code on their behalf has done the same. Vibe coding is essentially coding by describing what you want rather than writing syntax. So what’s the low-down on the state of things and where is the technology headed? Read on for my take.
Just a couple of short years ago, I had asked an LLM to help me edit an advanced script that I had written. The LLM broke my code, and I decided to add my desired add-on feature myself. That was then. This is now. Today, LLMs are much more powerful, and capable of understanding much more complex code thanks to larger context lengths and larger/smarter models. It is now realistic for anyone to ask an LLM to write code for them, regardless of their programming experience, and to have a working prototype (at a minimum) delivered to them soon thereafter. So how does this change programming for software developers?
If LLMs are writing the code, then what role do developers play, if any, in this new paradigm. Well, the “developer” role is going to shift from being the code-writing liaison for people (product managers), to being the people-liaison to the code writers (LLMs). Basically, developers will become more feature managers than coders and their scope of responsibility will be largely to prompt the LLMs to build the features, oversee the planning and coordination among disparate code blocks (I.e. feature co-architects), ensuring sensible efficient designs. They will oversee and ensure secure coding design principles are followed. Finally, they will conduct human usability, user experience, and feature testing (quality assurance).
Even with all of the developments in AI/LLMs recently, there are still some blind spots. For example, LLMs have the ability to “see” problems, but they still need human intervention to reproduce many of them. Also, LLMs can have add unnecessary complexity to code. Often , they violate Occum’s razor – the principle that simpler solutions are preferable—generating overly complex code when a straightforward approach would suffice
This is an exciting time for the software development field, and we are going to see rapid development of new innovative apps. Developers will shift from thinking about features in terms of “work”, and they will think about features in terms of “impact”. Since LLMs are doing the grunt work, focusing on the impact of new features is going to be the new top priority. Companies need to partner LLMs with creative minds in their organizations, not just developers, to best utilize these new coding capabilities. And for anyone worried about job displacement — as long as there are people using these apps, there will be people designing and testing them. The jobs will shift, but the explosion in productivity on the development side will open new opportunities for testers and feature managers. There will still be jobs!
