Open source is going to win it all: Harvey proves it | E2328
Спикеры
Jason Calacanis
— Ангельский инвестор / Основатель, LAUNCH
Lon Harris
— Co-host
Allan Guo
— Co-founder & CEO, Willow
Jason Calacanis and Lon Harris discuss the AI market, focusing on Harvey's in-house open-weight legal model as evidence that open source will win enterprise AI and cut frontier-model spend. They also cover OpenAI IPO chatter, Grokbot productivity agents, Willow's free AI dictation strategy, and a Stanford student's self-driving golf cart project. The conversation highlights accelerating autonomous vehicle development and strategic beneficiaries like Nvidia and Uber.
- Harvey released an in-house model built on open-weight Kimi K3, strengthening Jason's open-source AI thesis.
- Jason argues most corporate AI tokens will move off frontier models to internal/open-weight models.
- OpenAI's CFO reportedly targets a public listing by 2027 or sooner.
- Jason and Lon discuss Grokbot as a less brittle, always-running agent compared with earlier Claude/OpenClaw setups.
- Willow's founder argues free AI dictation will commoditize the market before Apple built-in tools reach near-parity.
- A Stanford student built a low-cost self-driving golf cart, leading Jason to raise his autonomous vehicle adoption outlook.
- Jason sees Nvidia and Uber as strategic beneficiaries of fragmented/open autonomous driving software.
Идеи
Open source AI wins, frontier models lose.
Jason argues that open source will win AI: the majority of corporate AI tokens will move off frontier models into internal/open-weight models because enterprises do not want to hand proprietary intelligence to labs that could compete with them. Harvey building on Kimi K3 and moving an estimated 99% of frontier spend is presented as evidence; frontier labs will be left with mostly distillation spend.
Open source AI wins, frontier models lose.
Jason argues that open source will win AI: the majority of corporate AI tokens will move off frontier models into internal/open-weight models because enterprises do not want to hand proprietary intelligence to labs that could compete with them. Harvey building on Kimi K3 and moving an estimated 99% of frontier spend is presented as evidence; frontier labs will be left with mostly distillation spend.
Proprietary data owners become AI winners.
As language models commoditize and open source wins, Jason believes the real winners are companies that own valuable expert/proprietary training data, such as legal AI and other application-layer companies. They can post-train open-weight models on their own data, creating specialized moats without sharing intelligence with frontier labs.
Autonomous vehicle adoption is accelerating quickly.
Jason believes autonomous vehicle technology is being figured out by far more players than he originally estimated. He previously expected 20 companies to solve autonomy in 24 months; seeing an undergrad build a self-driving golf cart in weeks makes him think it could be 2,000 players in 36 months, indicating accelerating AV adoption.
Uber benefits from open-source AV fragmentation.
Jason argues Uber wants open-source autonomous driving software to fragment the market so no single player such as Waymo, Pony AI, WeRide, or Zoox dominates; that fragmentation helps Uber preserve leverage and optionality across autonomous fleets.
Nvidia wins from AV software fragmentation.
Jason says Nvidia is strategically positioned to benefit from autonomous driving software becoming open and fragmented: Nvidia wants to sell compute hardware, and abstracted software plus many AV players means more edge compute like Jetson Thor is needed.
Далее
This This Week in Startups video, published August 21, 2026, features Jason Calacanis discussing Open source AI, Frontier AI models, Proprietary training data owners, Autonomous vehicles, UBER, NVDA. 6 trade ideas extracted by AI with direction and confidence scoring.
Speakers: Jason Calacanis · Tickers: Open source AI, Frontier AI models, Proprietary training data owners, Autonomous vehicles, UBER, NVDA