thereeldeel

"Garry Tan, CEO of Y Combinator, noted that by the Winter 2025 cohort, approximately 25% of the participating startups reported that 95% or more of their entire application codebase was generated by artificial intelligence". Curious to know what has happened since then. There would be some pros (fast to validate idea, pivot etc), some cons (code slop, poor architecture, etc). Would like to hear stories from those who have lived through this.

This question is inspired by the work done in https://transformer-circuits.pub/2025/attribution-graphs/biology.html . When we give a prompt such as https://blog.fsck.com/2025/10/05/how-im-using-coding-agents-in-september-2025/#:~:text=are%20the%20fixes%20they%20propose%20the%20correct%20ones , are we really expecting the llm to have built an adequate representation of 'whatever is the required world model' and reasoning to be able to answer if the 'PR is correct?' . Is any research on this?

Only Apache 2.0 license guarantees that the licensee doesn't get sued for patent infringement.

Why do majority of open source startups (licensors) use MIT license? Is it that that they want to keep the option of suing, on the basis of patent infringement, open?

So, to be safer no licensee should use code from a MIT licensed project, right?