I thought it was a good article and shared it. (Read the article at the link)
Summary: In order for AI to be properly utilized in the Vibe coding era, project proposals written by humans need to evolve into a 'structured system' that AI can also understand.
Core Problem: The Paradox of AI Coding
While the 'Vibe coding' era, where code is written in natural language, has opened up and accelerated development speed, paradoxically, code quality has decreased and review time has lengthened.
Phenomenon: Increase in copied code, sharp decrease in refactoring ratio, decline in trust in AI tools.
Cause: The problem is not with the performance of AI, but with the form of the project proposal. Documents written for humans are just fragmented 'unstructured text' for AI, so the AI cannot fully understand whether the planning intention is reflected in the code.
Solution: 'Vibe Planning'
The authors propose to address this by transforming project proposals into structured data that AI can understand.
Structuring: Hierarchical organization of the project proposal, such as Requirement → Feature → Spec → Userflow, to enable automatic comparison between code and planning by AI.
Visualization: Removing ambiguity through screen planning (wireframes) in addition to text.
Results: By having AI perform the manual project proposal-code comparison task, the time required for bug detection and correction was drastically reduced (e.g., from two days to two hours).
'Vibe Planning' Methods You Can Implement Right Now
Have AI read the project proposal: Feed the project proposal to AI and have it summarize the Feature list and priorities to test its understanding of the proposal.
Hierarchical Functioning: Break down vague planning like "login function" into detailed functions and exception cases.
Have AI read the plan before coding: Before coding, confirm that the AI has a proper understanding of the specifications it needs to implement.
▶ Original Source: https://eopla.net/magazines/43902?utm_source=copy_link&utm_medium=share#
