With the rapid advancement of artificial intelligence, more and more developers are relying on AI-generated code. This trend is particularly evident among the latest batch of startups at Y Combinator (YC), a renowned Silicon Valley startup accelerator. In a recent YouTube conversation, YC managing partner Jared Friedman revealed that a quarter of the Winter 2025 (W25) batch of startups have 95% of their codebase generated by AI.
Friedman clarified that this astonishing 95% figure excludes imported library code and refers specifically to the core code written collaboratively by humans and AI. He emphasized, "We're not funding a bunch of non-technical founders. These are highly technical people who are perfectly capable of building from scratch. A year ago, they would have. Now, 95% of the code is done by AI."
In a video titled "Ambient Coding is the Future," Friedman, along with YC CEO Garry Tan, managing partner Harj Taggar, and partner Diana Hu, discussed this trend. They noted that developers are shifting towards using natural language and intuition to write code, rather than the traditional line-by-line approach. Last month, Andrej Karpathy, former AI director at Tesla and former OpenAI researcher, coined the term "ambient coding" to describe this new way of coding with large language models (LLMs), where developers focus more on intent than code details.
However, AI-generated code is not without flaws. Numerous studies and reports indicate that AI-generated code can introduce security vulnerabilities, leading to application downtime or frequent errors, forcing developers to spend significant time debugging or modifying the code. Hu pointed out in the discussion that even with heavy AI reliance, developers still need a crucial skill: reading code and identifying errors. "You have to have taste, enough training to judge whether the LLM output is good or bad. To do 'ambient coding' well, you still need knowledge and discernment to distinguish between good and bad," she said.
Garry Tan agreed, adding that founders still need classical coding training to ensure the long-term stability of their products. "Suppose a startup with 95% AI-generated code successfully goes public and has 100 million users a year or two later. Will it crash? Current reasoning models aren't strong enough for debugging. So founders must have a deep understanding of the product," he advised.
The surge in AI coding has attracted significant attention from venture capitalists and developers. In the past 12 months, startups focused on AI coding, such as Bolt.new, Codeium, Cursor, Lovable, and Magic, have collectively raised hundreds of millions of dollars in funding. Tan commented, "This isn't a fleeting trend; it's the mainstream way of coding. If you don't keep up, you'll be left behind."
As AI models become increasingly integrated into coding, "ambient coding" is not only changing how developers work but also creating new possibilities for technology startups. However, finding the right balance between efficiency and quality remains a challenge developers must face as they work alongside AI.
