@chipro When there is no best strategy to find a “balance”, over design and over test were left to be the local optimization and equilibrium. Moreover, it burs more tokens.
@Jadu100x started with the excitement, ended up with desperately prompting the agent but it is still not moving a bit to converge to the goal…I literally had to restart the whole project 3 times to eventually make it to production.😢
Vibecoding builds fast, but after iterations, the repo became hard for human and agents to understand and maintain. (with gpt-5.6 Sol's infinity tests, it becomes nightmere...)
So I built VCRS: a practical standard for keeping AI-built repo sustainable: https://t.co/9suewstHmV