Startup Ideas You Can Now Build With AI

YC partners Garry, Harj, Jared, and Diana argue LLMs resurrect failed 2010s startup categories—recruiting, edtech, full-stack services—by replacing the human ops layer that killed margins.

• Harj’s Triple Bite raised ~$50M, competitor raised $100M+; entire recruiting marketplace category failed because human-run evaluation was unscalable—LLMs eliminate that bottleneck on day one.
• Meror (recruiting) and Apriora (technical screening) are direct 2025 rewrites of failed 2015 ideas, now viable because LLM evaluation replaces years of labeled-data accumulation.
• Full-stack startups (Atrium, WeWork-style) collapsed on gross margins; agents now let full-stack companies run ops at software margins for the first time.
• Legora (AI legal tools, fastest-growing YC company by their account) is the new Atrium—Justin Khan said the original idea was right, the 2018 AI just wasn’t good enough.
• Replicate (W20) stopped working on the product for months during COVID; exploded overnight when image diffusion models launched—timing on the oil well matters more than idea originality.
• Gemini 2.5 Pro matches or beats O3 on several tasks per internal YC evals, yet consumer usage is a negligible fraction of ChatGPT’s—brand moat is real and durable even against technically superior models.
• Classic lean-startup advice (customer discovery first, fail fast) is outdated for the AI era; follow technical curiosity instead, good ideas surface from using the technology.
• Hundreds of 100–1000-person unicorns have no internal AI transformation underway; the gap between AI capability and enterprise adoption is still enormous.

2025-05-16 · Watch on YouTube