25% of Y Combinator Startups Building Products with Almost Entirely AI-Generated Code
AI Opens a New Paradigm in Coding As AI models have dramatically improved their coding capabilities, cases of developers using AI to generate code are surging. Notably, in Silicon Valley's renowned startup accelerator Y Combinator (YC)'s latest W25 batch, 25% of startups generated 95% of their codebase with AI. This was directly mentioned by YC managing partner Jared Friedman in a YouTube conversation.
"These aren't non-technical founders. They're all highly technical people who, a year ago, would have built their products from scratch. But now 95% is being built by AI."
- Jared Friedman
The Rise of 'Vibe Coding': A New Way to Code
Friedman, along with YC CEO Garry Tan, managing partner Harj Taggar, and general partner Diana Hu, discussed AI coding trends. They talked about 'Vibe Coding' — a new approach to generating code using natural language and intuition.
The term was first coined by Andrej Karpathy, former Tesla AI lead and ex-OpenAI researcher. He proposed the term to describe coding with large language models (LLMs) without focusing on the code itself.
"Vibe Coding isn't just about writing code — it's about intuitively creating code in collaboration with AI."
- Andrej Karpathy
The Limitations of AI Coding and the Developer's Role
However, AI-generated code isn't perfect. Research and reports indicate that AI-generated code can introduce security vulnerabilities, cause application failures, or make mistakes. Developers must still invest significant time in code review and debugging.
Diana Hu emphasized that code reading and bug-finding skills remain important even when coding with AI.
"To do good 'Vibe Coding,' you need the taste and knowledge to judge whether the LLM's output is good or bad."
- Diana Hu
Garry Tan added that foundational coding training is essential for AI coding to be sustainable long-term.
"Imagine a product made with AI-generated code goes to market and has 100 million users in a year or two. What happens if the product doesn't work properly? Early AI models aren't great at debugging, so you need a deep understanding of how the product works."
- Garry Tan
The Future of AI Coding: AI Has Become Mainstream
Interest from venture capitalists (VCs) and developers in AI coding tools is intense. In recent months, startups like Bolt.new, Codeium, Cursor, Lovable, and Magic have raised hundreds of millions of dollars. They're focused on revolutionizing coding through AI.
Garry Tan expressed conviction that AI coding isn't a passing fad but will become the dominant way of coding.
"This isn't just a fad. It's not going away. This is going to become the dominant way of coding. If you're not doing this, you might fall behind."
- Garry Tan
Key Terms Summary
- Y Combinator (W25 batch): 25% of startups generated 95% of code with AI
- Vibe Coding: AI-based coding approach using natural language and intuition
- LLM (Large Language Models): Core technology behind AI coding
- AI coding limitations: Security vulnerabilities, failures, debugging issues
- Developer's role: Code review, debugging, foundational coding knowledge essential
- AI coding startups: Bolt.new, Codeium, Cursor, Lovable, Magic, etc.
AI is changing the landscape of coding right now, with developers writing code in new ways through collaboration with AI. But the importance of human judgment and skill remains — something we shouldn't forget