LLMs are perfectly capable of writing correct code
My understanding is that this is only partly true.
The problem as I understand it is that an LLM can create a code base for a prompt like “create an android app where I can record my medications”, and it might be “correct” and surprisingly usable. However, as your app evolves and “develops” it will be a mess, and the cognitive load quickly becomes unbearable. Everything feature breaks a heap of stuff and every fix breaks more things.
It depends on how it’s built. Anyone vibecoding without a plan will run into this very quickly. However, if a project is planned out in advance, broken down into multiple milestones, with measurable goals and thorough testing at each milestone before moving forward, the LLM can create surprisingly well-structured and solid code which fits the overall structure.
Basically, the more thought, planning, and structure is worked out before anybody runs even a single prompt, the better your results will be. And continual testing with the goal of finding every single weak point will give even better results. But planning it out ahead of time and doing all that testing takes a lot of time and human effort, and nobody vibecoding slop apps is doing that.
My understanding is that this is only partly true.
The problem as I understand it is that an LLM can create a code base for a prompt like “create an android app where I can record my medications”, and it might be “correct” and surprisingly usable. However, as your app evolves and “develops” it will be a mess, and the cognitive load quickly becomes unbearable. Everything feature breaks a heap of stuff and every fix breaks more things.
It depends on how it’s built. Anyone vibecoding without a plan will run into this very quickly. However, if a project is planned out in advance, broken down into multiple milestones, with measurable goals and thorough testing at each milestone before moving forward, the LLM can create surprisingly well-structured and solid code which fits the overall structure.
Basically, the more thought, planning, and structure is worked out before anybody runs even a single prompt, the better your results will be. And continual testing with the goal of finding every single weak point will give even better results. But planning it out ahead of time and doing all that testing takes a lot of time and human effort, and nobody vibecoding slop apps is doing that.
That’s right. I think vibecoding generally refers to people who don’t understand code and don’t have the ability to plan around potential problems.