Ioannis Antonoglou, CTO & Co-Founder, Reflection AI
At the Sierra Ventures 20th Annual CXO Summit, Ioannis Antonoglou, co-founder and CTO of Reflection AI and one of DeepMind's founding engineers, explored one of the defining questions of this era: who will control the intelligence that shapes the modern world?
His perspective centered on the urgent need to build powerful, open foundation models in the U.S. and to pair them with new advances in reinforcement learning that can move AI from reasoning to real agency.
Keeping Intelligence Open
Open models have historically played the same role for AI that open software did for the internet, forming a foundation for transparency, innovation, and shared progress. Without powerful open-weight systems that researchers and developers can freely access and adapt, AI risks becoming locked behind the walls of a few proprietary providers. The next generation of breakthroughs will likely depend on a healthy, open ecosystem where ideas can move freely, and advances can be built upon collaboratively.
From Reasoning to Agency
The current phase of large language models excels at reasoning and recall, but real progress toward intelligence requires systems that can act, learn, and improve autonomously. Reinforcement learning provides this missing layer. By integrating it with language models, AI can move beyond static prediction into decision-making and iterative problem-solving, laying the groundwork for true agency and more general intelligence.
A New Race for Open Source Leadership
Some of the most capable open-weight models today are being developed in China, reflecting a strategic understanding that the ecosystem layer of AI, the models others build upon, will define long-term leadership. For the U.S., building its own frontier open models is both a technological and geopolitical imperative. Open systems represent more than competitiveness; they embody values of transparency, accessibility, and collaboration that align with the principles underpinning the digital economy.
Building Balance in the AI Frontier
The challenge ahead is not just to match the capabilities of closed systems but to ensure that open models remain at the frontier. The goal is a balanced landscape where open and closed systems coexist, driving progress through diversity rather than consolidation. Sustaining that balance will determine whether intelligence remains a shared global resource or becomes concentrated in a few private hands.
A Shared Responsibility
Across the CXO Summit, a consistent theme emerged: responsible innovation requires openness, trust, and access. As enterprises adopt AI at scale, the ability to build on transparent and verifiable foundations may prove to be America’s most enduring advantage.
