With the release of new models of the ones that are best in coding, developers are increasingly using it to generate code. One of the newest examples is the current group of Y Combinator, the accelerator of the start of Silicon Valley. A quarter of the W25 start series have 95% of the basics of their codes created by him, said YC Jared Friedman’s management partner during a conversation posted on YouTube.
Friedman said this 95% figure does not include things like the written code to import libraries, but took into account the code printed by people compared to it.
“It’s not how to fund a bunch of non -technical founders. Each of these people is very technical, fully capable of building their products from scratch. A year ago, they would have built their product from scratch – but now 95% of it is built by one,” he said.
In a video titled “Vibe Coding is the Future”, Friedman, along with YC Garry Tan’s CEO, Harj Taggar management partner and Diana Hu partner, discussed the trend of natural language and instincts to create code.
Last month, the former head of him in Tesla and former Openai, Andrej Karpathy described the term “Vibe coding” to describe a way to encode using large linguistic models (LLM) without focusing on the code itself.
The code generated by it is far from the perfect, through. Studies and reports have noted that some codes created by it can enter the security flaws into applications, cause interruptions or make mistakes, force Devs to change the code or debug too much.
During the discussion, Hu said that even if the product builders rely heavily on him, one skill that should be good is reading the code and finding mistakes.
“You need to have enough taste and training to know that a llm is spitting bad things or good things. To do ‘good coding’, you still have to taste and knowledge to judge well against bad,” she said.
Tan also agreed on the point of founders in need of classical coding training to keep products in the long run.
“Let’s say that a 95% code generated by it comes out (on the market), and one year or two outside, they have 100 million users in that product, whether it falls? The first versions of reasoning patterns are not good in debugging. So you have to go to the depths of what is happening with the product,”
VCS and developers are harassed over the coding of AI. Beginnings including Bolt.new, Codeium, Cursor, Lovable and Magic have collected hundreds of millions of dollars in funds in the last 12 months.
“This is not a whim. That’s not going away. This is the prevailing way to encode. And if you are not doing it, you can simply be left behind,” Tan added.
