Summary
Harry Stebbings of 20VC sits down with Eno Reyes, CTO and co-founder of Factory — one of the leading autonomous software development companies — for a wide-ranging conversation on how enterprises should think about AI model selection, open-source Chinese models, and the economics of intelligent systems.
Reyes makes the case that evaluating models by token price is the wrong unit of analysis: what matters is cost per correct outcome, not cost per token. Under that framing, a more capable model that completes a code review in fewer passes can be cheaper end-to-end than a budget model that burns tokens iterating toward the same answer. He also pushes back hard on the framing of open-source Chinese models as inherently risky, calling it a deliberate narrative by frontier labs to protect their competitive moats. His prediction: within three years, 99% of AI workflows will run on open models.
The conversation also covers Factory’s on-premise offering, sovereign intelligence as an enterprise concern, and the implications of Cursor’s acquisition by SpaceX — which Reyes argues will push Cursor toward Grok model lock-in, creating an opening for model-agnostic competitors like Factory. The episode includes discussion of the broader neo-lab landscape, with Reyes suggesting that only around 10% of new AI labs ultimately survive. A dense, analytically rigorous episode for anyone tracking the enterprise AI competitive landscape.
📺 Source: 20VC with Harry Stebbings · Published August 29, 2026
🏷️ Format: Podcast
