Y CombinatorThe Truth About Building AI Startups Today
Jared Friedman on yC Partners Reveal How To Build Durable, Billion-Dollar AI Startups.
In this episode of Y Combinator, featuring Jared Friedman and Garry Tan, The Truth About Building AI Startups Today explores yC Partners Reveal How To Build Durable, Billion-Dollar AI Startups YC group partners discuss why so many top founders are pursuing AI and how this moment resembles an unprecedented gold rush of startup opportunities. They emphasize that the best AI companies focus on specific, boring workflows and deep vertical problems rather than generic chatbots or shiny “GPT wrapper” ideas. The conversation covers AI tarpit ideas, fine-tuning and open-source models, LLM security, and how to avoid getting steamrolled by future foundation models like GPT‑5. Overall, they argue that this era re-centers hardcore technologists and offers once-in-a-lifetime chances for founders who move fast and solve concrete problems.
At a glance
WHAT IT’S REALLY ABOUT
YC Partners Reveal How To Build Durable, Billion-Dollar AI Startups
- YC group partners discuss why so many top founders are pursuing AI and how this moment resembles an unprecedented gold rush of startup opportunities. They emphasize that the best AI companies focus on specific, boring workflows and deep vertical problems rather than generic chatbots or shiny “GPT wrapper” ideas. The conversation covers AI tarpit ideas, fine-tuning and open-source models, LLM security, and how to avoid getting steamrolled by future foundation models like GPT‑5. Overall, they argue that this era re-centers hardcore technologists and offers once-in-a-lifetime chances for founders who move fast and solve concrete problems.
IDEAS WORTH REMEMBERING
7 ideasChase boring, specific workflows instead of flashy generic AI ideas.
The most promising AI startups automate mundane, information-processing tasks (e.g., government contract bidding) where humans currently read, summarize, and re-enter data—these are perfect fits for LLMs and face little competition.
Avoid AI “tarpit” ideas that attract many founders but lack real usage.
Concepts like generic AI co-pilots or vague ‘throw your data in and we’ll automate everything’ tools are easy to sell and pre-sell, but hard to make truly useful because customers don’t know concrete use cases.
Think beyond chat interfaces; embed LLMs into familiar, task-focused UIs.
Relying solely on chat forces users to know how to ‘talk to a computer’; instead, use LLMs behind the scenes in traditional web or mobile interfaces that align with existing workflows and jobs-to-be-done.
Compete on better outcomes, not just cheaper models.
Fine-tuning open-source models only for cost savings is fragile because foundation model prices keep dropping; lasting value comes from domain customization, proprietary data, and superior performance on specific tasks.
Leverage smaller, domain-specific models where they outperform general LLMs.
For domains like coding, hardware, or SQL parsing, older or smaller models fine-tuned on narrow vocabularies can be “good enough” or better, cheaper, and faster than state-of-the-art general models.
Expect and exploit new ecosystems: dev tools, security, governance for LLMs.
Just as cloud created entire cybersecurity and tooling industries, LLMs are spawning needs like prompt security, data leakage prevention, access control, and local inference tooling—each a potential company.
Anchor your AI startup in a clear, defensible vertical use case.
Ideas most at risk of being ‘run over by GPT‑5’ are overly general; founders should deeply encode business logic, regulations, and workflows of a specific sector (e.g., healthcare compliance, government forms, SMB voice agents) where generic models can’t easily substitute.
WORDS WORTH SAVING
5 quotesWhere there's muck, there's brass.
— Garry (quoting Paul Graham / old saying)
Many of the companies get into YC… and within a month after we fund them, they're looking for a new idea… but man, was it easy last summer. There were great startup ideas just lying on the ground.
— Jared (YC partner)
All of SaaS software is just MySQL wrappers.
— Garry
We actually want some form of equity at the AI level… not merely the biggest companies to own the most capable AIs.
— Jared
This might actually be a once-in-a-lifetime opportunity… and I think I actually agree.
— Diana
QUESTIONS ANSWERED IN THIS EPISODE
5 questionsHow can a new founder systematically uncover ‘boring’ but high-value workflows in specific industries that are ripe for AI automation?
YC group partners discuss why so many top founders are pursuing AI and how this moment resembles an unprecedented gold rush of startup opportunities. They emphasize that the best AI companies focus on specific, boring workflows and deep vertical problems rather than generic chatbots or shiny “GPT wrapper” ideas. The conversation covers AI tarpit ideas, fine-tuning and open-source models, LLM security, and how to avoid getting steamrolled by future foundation models like GPT‑5. Overall, they argue that this era re-centers hardcore technologists and offers once-in-a-lifetime chances for founders who move fast and solve concrete problems.
What concrete signals distinguish an AI tarpit idea from an early but promising concept before you’ve spent months building?
Where is the line between a useful domain-specific co-pilot and a generic chatbot that will inevitably be subsumed by foundation model providers?
How should startups balance building on closed foundation models vs. investing early in their own domain-specific or open-source models?
What governance, security, and access-control layers around LLMs will become table stakes for enterprises over the next 3–5 years?
EVERY SPOKEN WORD
Install uListen for AI-powered chat & search across the full episode — Get Full Transcript
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome