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

With the release of new AI models that are better during coding, developers use AI to create a code. One of the latest examples is the current group of the ConnoBinator of the Silicon Valley Starting Accelerator. A quarter of the W25 Startup Team has 95% of the codes created by AI, YC-controlled partner Jared Friedman A conversation sent on YouTube.

Friedman, this 95% digit does not include things like the code written to import into libraries, but did not take into account the code written by people compared to AI.

“This is not as we have financed by a group of non-technical founders. Each of these people can completely build its products from scratch. But now 95% are set by an AI,” he said.

In a video called “Vibe Coding”, Friedman, YC CEO Garry Tan, management partner Harge Taggar and general partner Diana HU, discussed the use of natural language and instincts to create a code.

Last month, the former head of the EU Tesla and Ex-researcher Openai, Andrej Karpathy described the term “vibe coding” To describe a way to use large language models (LLMS) without taking care of the code.

The code generated from the AI ​​is perfect, very far away. Research and reports observed that Some AI Created Code can include security defects in applications, cause to cause cutor make a mistakeTo change the code or force devs to debug.

During the discussions, HU said, although the product builders are very confident in the AI, he said that they have to read a skill code and find mistakes.

“You need to train and exercise enough to know that LLM’s bad things or good things are spit. Good and bad and bad vs should be pleasant and knowledge to do badly and bad.”

Tan also agreed with the points of the founders in need of classical coding training for the products in a long time.

“Let’s say 95% EU is a start with a coded code exit [in the market]And for one or two years, there are 100 million users in this product, fall on top or not? The first versions of the models are not good in the separation. Thus, you must go to the depth of what happened with the product. “

VCS and developers were excited AI-Powered Coding. Beginnings including Bolt.new, Codey, Cursor, Cuteand Magic Funded hundreds of millions of dollars in the last 12 months.

“It’s not fad. It doesn’t go. This is the dominant way of this code. You can simply stay behind, if not.”