A quarter of startups in YC’s current cohort have codebases that are almost entirely AI-generated

With the publication of better new AI models in coding, developers are using growing AI to create codes. One of the new examples is the current batch of Wi -combinator, the Store Silicon Valley Startup Accelerator. A quarter of the 25 startup batch of the BULE has 95% of their codebus produced by AI, the YC Managing Partner Jared Fredman says A conversation posted on YouTubeThe

Fredman said that this 95% of the figure did not include things like written code for importing the library but took into consideration the code typed by humans compared to AI.

“It is not that we have financed a number of non-technical founders. Each of these people is extremely technical, completely capable of making their own products from scratches. A year ago, they could make their product from scratch – but now 95% of it was built by AI, “he said.

In a video titled “Vibe Coding is Future”, Fredman discusses the trend of using natural language and instinct to create codes including WAC CEO Gary Tan, Managing Partner Herz Tagger and General Partner Diana Hu.

Last month, former AI chief of AI and former researcher at OpenAI, Andrez Carpothy Described the word “siblings coding” To describe any way of the code using the big language model (LLM) without focusing on the code itself.

The code generated from AI is far from perfect. Have observed the study and reports Some AI-exposed codes may insert security errors in applications, Cause outageOr Make the mistakeDaves to change the code or forced to make heavy debugs.

During the discussion, Hu said that the product manufacturers depend a lot on AI, but they need to be good for reading the code and looking for bugs.

“You need to have enough taste and adequate training to find out which LLM is giving bad things or good things. For good ‘coding’, you still need to have taste and knowledge to judge the good vs. bad, “he said.

Tan has also agreed to the point of the company’s point in order to maintain the products in the long run.

“Let’s say a startup with 95% AI-exposed code [in the market]And out of one or two years, they have 100 million users on that product, does it not fall? First version of reasonable models is not good in debugging. So you have to go to the depth of what’s happening to the product, ”he advised.

Have been encouraged about the VC and developers AI-driven codingThe Startups with Bolt.New, Codium, Cursor, LovingAnd Mud Has raised several million dollars of funds in the last 12 months.

“This is not a fad. It’s not going away. This is the dominant way of the code. And if you haven’t done it you can only stay behind, “added the tension.