Subvolts take Autonomous coding agents are coming, and a flexible approach beats tightly coupled AI. Jul 21, 2026 - Developers
- Learn about Factory's contrarian technical approach to building autonomous coding agents and the future of software development.
- Business
- Understand the economic arguments for open-weight models and how AI token costs will be justified like headcount.
This episode provides a deep dive into autonomous coding agents from a startup that has been building them for years.
Factory's Matan Grinberg: The Coming ‘Dark Factory’ Where Software Builds Itself
Audio streams from Sequoia Capital. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
Sequoia Capital's Training Data podcast features Matan Grinberg, CEO of Factory, discussing the development of autonomous coding agents. Grinberg shares Factory's "journey in the desert," their approach to model-agnostic harnesses, and predicts a future "dark factory" where software builds itself asynchronously.
Imagine software that writes itself. Matan Grinberg talks about building these "autonomous coding agents." He believes a flexible system that works with many AI models, instead of just one, will be more successful and cheaper.
This episode offers a contrarian technical perspective on AI development, particularly for those interested in the future of software engineering. Grinberg's insights on model-agnostic systems and the economics of open-weight models are valuable for understanding the evolving AI landscape beyond frontier labs. It's for developers and business leaders questioning the path of large, integrated AI systems and seeking alternative strategies for cost-effective AI adoption.
- HostSequoia Capital
- GuestMatan GrinbergCEO, Factory
- Guest: Matan Grinberg, CEO of Factory
- Show: Training Data (Sequoia Capital)
- Length: 52 minutes
- Factory started building autonomous coding agents: April 2023
- Prediction for coding tokens: 90% will run asynchronously
Factory started building fully autonomous coding agents in April 2023, two years before enterprises were ready. Matan Grinberg now says this is indistinguishable from being wrong. The Factory co-founder and CEO explains how the company survived its "journey in the desert," including the decision to hand nearly all of its revenue back to customers when the product wasn't making developers obsessed. Matan makes the contrarian technical case that a model-agnostic harness beats the model-and-harness co-design that labs like OpenAI and Anthropic favor, because exposing a harness to many models keeps it from overfitting to any single one. He argues open-weight models like GLM will capture the majority of tokens by staying one generation behind the frontier at a fraction of the cost, and that CIOs will soon justify every incremental token the way they justify headcount. Looking ahead, he predicts 90% of coding tokens will run asynchronously—the "dark factory" where software builds itself.
Hosted by Sonya Huang and Pat Grady, Sequoia Capital
- autonomous coding agents
- Matan Grinberg
- AI systems that can write software code on their own.
- model-agnostic harness
- Matan Grinberg
- A flexible system that can work with many different AI models without being tied to just one.
- overfitting
- Matan Grinberg
- When an AI model becomes too specialized to one type of data and performs poorly on new, different data.
- open-weight models
- Matan Grinberg
- AI models whose underlying code and data structures are publicly available, allowing others to build upon them.
1 h 08David Silver and Lasse Espeholt of Ineffable Intelligence: A Superlearner for the Age of Experience
WithDavid Silver, Lasse Espeholt
1 h 041 h 0552 minMaking Cities Awesome: Peregrine’s Nick Noone & Ben Rudolph
WithNick Noone, Ben Rudolph
21 minHow to Get the Most from AI This Summer
WithEthan Mollick
1 h 05
2 h 34The Hugging Face Breach, Moonshot AI Valued at $20B, and Living to 1,759 Years Old | EP #273
WithSalim Ismail, Dave Blundin, Alexander Wissner-Gross
Questions people ask
- What is Factory's approach to AI models?
- Factory uses a model-agnostic harness, which they believe prevents overfitting by exposing it to many different models, contrasting with the co-design favored by labs like OpenAI and Anthropic.
- What is the 'dark factory' concept?
- The 'dark factory' is Grinberg's prediction that 90% of coding tokens will run asynchronously, meaning software will largely build itself without direct human intervention.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Sequoia Capital.