Will the agent era be built on an open-source stack or on proprietary platforms?

Ethan Mollick · Jul 31, 2024

A freely downloadable GPT-4-class open-weight model is a big deal that will unlock weird AI effects that had been delayed, because until now every national government had worse AI than any kid with access to GPT-4o

The capability was previously gated behind labs' hosted products; an openly available fine-tunable model puts frontier capability into everyone's hands, including actors who could not access it before

5:36 20VC: Is More Compute the Answer to Model Performance | Why OpenAI Abandons Products, The Biggest Opportunities They Have Not Taken & Analysing Their Race for AGI | What Companies, AI Labs and Startups Get Wrong About AI with Ethan Mollick

Ethan Mollick · Jul 31, 2024

Open AI models will do substantial good — sparking entrepreneurship and delivering low-hanging-fruit gains in healthcare and education in places without access to good doctors or tutors

He is generally pro technological progress and openness frees people to do interesting things; evidence that founders in Kenya who get AI advice do better, at least if they were already doing well

Scope: benefits are clearest in regions lacking good doctors or tutors; in wealthy contexts the case for turning to AI is more nuanced; Kenya result holds for founders already performing well

17:48 20VC: Is More Compute the Answer to Model Performance | Why OpenAI Abandons Products, The Biggest Opportunities They Have Not Taken & Analysing Their Race for AGI | What Companies, AI Labs and Startups Get Wrong About AI with Ethan Mollick

David Cahn · Aug 5, 2024

Meta's open-source model strategy is the right approach — good for the world, good for Meta, and good for startups.

Startups will build on top of Llama and open source, and Meta's distribution means that if it unlocks the consumer killer use case it will monetize heavily.

26:52 20VC: Sequoia's David Cahn on AI's $600BN Question | Why the Data Centre is the Most Important Asset | Servers, Steel and Power: The Core Pillars Powering the Future of AI

Ryan Petersen · Jun 20, 2026

Open source commoditizing frontier models would be a good outcome, not something to fear

If AI and intelligence become really cheap that's much better for the world and for him personally, and his Anthropic position isn't large enough to matter

Scope: speaking from his personal financial position

28:21 20VC: Why Remote Work is White Collar Fraud | Why Revenge and Patriotism are the Best Founder Traits | Two Questions Every Founder Needs to Ask | The Wild Story of Raising $1BN from Masa Son in an Hour Long Meeting with Ryan Peterson, Founder @ Flexport

Jeff Seibert · Nov 22, 2023 · hedged

The market will converge on a common, popular open-source base LLM plus tooling that makes fine-tuning easy, like a default Linux everyone customizes for their own product and market

Training an LLM takes enormous time, data, energy and money, but fine-tuning one can be done straightforwardly, so the base layer becomes shared infrastructure and differentiation moves to customization

Scope: framed as 'my sense'

23:41 20VC: Why OpenAI Will Become an Infrastructure Play, Why Apple Will Win in an AI World, Why Google is the Most Vulnerable Incumbent, Will LLMs Be Commoditised, Which Startups Are Thin vs Thick Wrappers on Top of LLMs with Jeff Seibert, Founder @ Digits

Jerry Murdock · Feb 28, 2026

An open-source 'claw stack' for autonomous agents will emerge, analogous to how the LAMP stack unlocked the 2004 web and commerce explosion

In 2003-04 expensive Sun servers and Oracle databases blocked website building until Linux/Apache/MySQL/PHP made it cheap, which triggered the explosion Google rode; the same open-source stack dynamic will repeat for agents

8:14 20VC: Why Cursor is Dead | An AI Tsunami is Coming & You Need to Prepare | Systems of Record Become Valueless Databases with Agents | Is This The End of Tech Private Equity with Jerry Murdock, Co-Founder of Insight Partners

Arthur Mensch · Apr 29, 2024

Bringing freedom to developers to own and modify models is the best way to distribute generative AI as widely as possible

When Mistral started the field looked like it would centralize around a couple of API providers; open source let the community realize they could build interesting technology by modifying models rather than depending on a few providers

Scope: open source part described as 'I believe, a good enabler'

17:02 20VC: Mistral's Arthur Mensch: Are Foundation Models Commoditising | How Do We Solve the Problem of Compute | Is There Value in the Application Layer | Open vs Closed: Who Wins and Mistral's Position

Arthur Mensch · Apr 29, 2024

Distribution is very important in enterprise AI, and releasing open-source models is a shortcut to distribution by creating demand

Open models create bottom-up demand that substitutes for incumbent distribution advantages

27:50 20VC: Mistral's Arthur Mensch: Are Foundation Models Commoditising | How Do We Solve the Problem of Compute | Is There Value in the Application Layer | Open vs Closed: Who Wins and Mistral's Position

Joelle Pineau · Nov 3, 2025

The industry's move toward closed models and closed access is a deep mistake and will not be effective — ideas circulate regardless, and closing down harms innovation

Research requires ideas to circulate; the belief that you can lock knowledge down is false because people and ideas move, and closure undermines the innovation ecosystem

54:51 Closing access fails because ideas and people circulate anyway

Christian Lanng · Sep 27, 2023

AI development has shifted from institutional 'cathedrals' like Google doing serious science to a distributed underground of people hacking on LLM models in Discord, and this is a positive, energising change

Five or six years ago AI was made by serious scientists inside big labs; now the building happens in open communities, which feels unlike anything he's been part of

63:59 Innovation migrated to distributed open communities

Amjad Masad · Apr 25, 2026

Strong open-source models are essential to keeping AI a free market; an oligopoly of AI labs would collude on price, restrict how AI is used, and withhold the best models from APIs

Economic theory predicts natural price collusion in an oligopoly, and it would be bad for AGI to be controlled by a few corporations

16:03 Open source is the only check on lab oligopoly collusion