The Twenty Minute VCGeorge Sivulka, Co-Founder & CEO @Hebbia: The Future of Foundation Models | E1250
At a glance
WHAT IT’S REALLY ABOUT
Hebbia’s George Sivulka: Beyond RAG Toward Agentic, Enterprise-Grade AI
- George Sivulka, founder and CEO of Hebbia, traces his path from misfit math kid with immigrant-athlete parents to Stanford PhD dropout and AI founder living in a literal closet while building the company.
- He argues that early-life adversity often fuels great founders, and that relentless persistence, almost to unhealthy extremes, is a core ingredient in startup success.
- On the product and industry side, Sivulka contends that Retrieval-Augmented Generation (RAG) mostly fails in real enterprise use, and that the future lies in agentic systems, platform-style “AI Excel” products, and scaling compute at inference rather than just bigger training runs.
- He predicts $100 trillion of new economic value from AI, believes model providers will commoditize while infrastructure and application layers win, and sees xAI, AMD, and agentic platforms like Hebbia as major beneficiaries.
IDEAS WORTH REMEMBERING
5 ideasEarly adversity can be a powerful founder motivator.
Sivulka and Stebbings both connect feeling like family misfits or disappointments to a deep drive to prove themselves, arguing many iconic founders share backgrounds of trauma, queerness, or adoption that create an enduring chip on the shoulder.
Relentless persistence often matters more than immediate fit or talent.
His NASA internship story—getting rejected five times, showing up in person, cold-calling from the sidewalk, bombing the interview, then returning with overnight self-study—illustrates how refusal to give up can brute-force opportunities that credentials alone wouldn’t.
RAG is overhyped and fails most real enterprise questions.
Hebbia deployed RAG at large finance firms and found ~90% of queries weren’t simple ‘find this quote’ tasks; they required reasoning about documents, not just retrieving passages, leading him to call most enterprise RAG deployments “vapor” that demo well but fail in production.
The next frontier is agentic systems that scale compute at inference.
Instead of only training ever-bigger models, Hebbia orchestrates hundreds or thousands of LLM calls across documents for a single question, effectively assembling many smaller ‘engines’ at inference time to achieve higher accuracy on complex tasks like deal diligence.
AI platforms will coexist with specialized apps and agents, not replace them.
Using Excel’s history as an analogy, Sivulka argues that horizontal platforms (like Hebbia) and vertical applications will mutually reinforce each other, with ‘agent employees’ emerging as a new layer rather than all business software collapsing into a single agent.
WORDS WORTH SAVING
5 quotesYou can brute force your way as a founder. Just screw product–market fit; you could literally brute force anything in the world.
— George Sivulka
I actually don’t think RAG works at all.
— George Sivulka
If there was $100 trillion of value created from the introduction of the computer, I actually think $100 trillion of value will be created from the introduction of AI compute in the next 60 years.
— George Sivulka
Chat is a useful interface, but it’s like a single cell in Excel.
— George Sivulka
Tech is not the hard part of all of this… the hardest part of AI change management, no matter what company you are, is people.
— George Sivulka
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