Skip to content
YC Root AccessYC Root Access

Lessons from Building Open Source Libraries

During last month’s NeurIPS 2025 conference, YC’s Diana Hu sat down with Thomas Wolf, co-founder and CSO of Hugging Face to discuss his unconventional journey from physics and law to building one of the most influential open-source AI platforms. They discussed why open research accelerates innovation, the real challenges of turning AI demos into products, and how great open models and the application layer unlock the biggest opportunities for founders. Apply to Y Combinator: https://www.ycombinator.com/apply Work at a startup: https://www.ycombinator.com/jobs Chapters: 00:00 — From Physicist to Hugging Face Founder 01:50 — Switching Careers 02:45 — How Hugging Face Was Born (Almost by Accident) 04:50 — The Limits of Closed Models 05:45 — Why Demos Often Don’t Become Real Products 07:05 — Fine-Tuning vs. Scaffolding: Startup Tradeoffs 08:40 — Turning Research into Widely Used Products 09:50 — Designing Great Developer Experiences 11:55 — The Future: Open Models and the App Layer

Diana HuhostThomas Wolfguest
Jan 15, 202614mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

How Hugging Face Turned Open Source Libraries Into Platforms

  1. Wolf’s career shifts (physics to law to startups) shaped Hugging Face’s mix of deep technical exploration and disciplined time/value prioritization.
  2. Hugging Face emerged accidentally from a game startup when an internal deep-learning exploration became a viral open-source library, prompting a pivot to an open-science platform mission.
  3. Open source accelerates AI progress by enabling direct code/model reuse, fast iteration, and creative recombination of strong pretrained bases into new domains and products.
  4. Most impressive AI demos fail to become products because real-world deployment demands domain knowledge, edge-case handling, and “scaffolding” or fine-tuning beyond out-of-the-box model behavior.
  5. As open models approach closed-model capability, competitive advantage shifts toward developer experience, user interaction design, and application-layer integration rather than model training alone.

IDEAS WORTH REMEMBERING

5 ideas

Career diversity can translate into better product and execution instincts.

Wolf credits physics with teaching deep exploration and law with forcing rigorous time valuation—together informing how to build impactful tools without losing practicality.

Open source wins by making iteration cheap and recombination easy.

When code and models are open, teams can tweak existing systems (not reinvent them) and quickly test new ideas by building on massive pretrained “base layers.”

Closed models constrain use cases; open access enables domain adaptation.

APIs like ChatGPT tend to work best within intended use patterns; having weights/code lets teams fine-tune or modify models for niche DSLs or unfamiliar domains.

The demo-to-production gap is mostly about reliability and edge cases.

Regardless of open or closed models, real products require domain expertise, preprocessing, guardrails, and operational “scaffolding” to meet user expectations in messy environments.

Fine-tuning is powerful but may be the wrong early-stage startup spend.

Startups must decide whether to allocate scarce time to training/fine-tuning (especially if the capability doesn’t exist otherwise) or rely on scaffolding and improving base models/tools.

WORDS WORTH SAVING

5 quotes

I guess one of the thing for me was always to try to work with people I wanted to work with even more than what I was specifically working on.

— Thomas Wolf

And then I just think life is too short to do just one thing, right?

— Thomas Wolf

Open source is probably the best thing that computer science brought to humanity.

— Thomas Wolf

But yeah, I think in any case, there is no super shortcut from demo to production. It's still a painful process.

— Thomas Wolf

No user of any software want to read the documentation. So, uh, it shouldn't even have to write the documentation. So everything should look really obvious.

— Thomas Wolf

Career pivots shaping founder mindsetHugging Face origin and missionOpen-source advantages and constraintsWhy demos don’t become production productsScaffolding vs. fine-tuning tradeoffsDesigning developer onboarding and abstractionsOpen models catching up; value moving to the app layer

High quality AI-generated summary created from speaker-labeled 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