Amazon is betting on agents to win the AI race

In this episode of Decoder, Alex Heath interviews David Luan, head of Amazon's AGI Research Lab, about the current state and future direction of AI, particularly focusing on the development of AI agents. Luan discusses his background at OpenAI and Adept, his current work at Amazon, and his vision for AGI as a universal teammate for knowledge workers. The conversation covers the limitations of current LLMs, the importance of training agents through self-play in simulated environments, and Amazon's strategic approach to AGI, including the use of agents within its various operations and the development of NovaACT. Luan also shares his insights on the talent market in AI and offers advice for those considering a career in the field.

Outlines

Sign in to continue reading, translating and more.

Continue

Decoder with Nilay Patel

Introduction to AI Agents and David Luan's Background

David Luan's Journey in AI and Reflections on GPT-5

The Maturing AI Landscape and the Focus on Practical Applications

Defining AGI and Amazon's Strategic Vision

The Current State and Future Promise of AI Agents

Overcoming Limitations of LLMs and Training Agents for Real-World Decision-Making

Amazon's Approach to Training AI Agents: Large-Scale Self-Play

The Future of AI Agents and the Importance of Product Form Factors

Amazon's Internal Collaboration and the Strategic Advantage of Private Data

Internal Adoption of Agents within Amazon and the Focus on Reliability

The Timeline for Advanced AI Agents and Amazon's Leapfrog Strategy

The Talent Market and the Reverse Acquihire Phenomenon

The Rising Costs of AI and the Importance of Partnering with a Major Player

The Limited Pool of AI Experts and Advice for Navigating the Field

Conclusion and Episode Wrap-up

Amazon is betting on agents to win the AI race

Decoder with Nilay Patel

00:00Introduction to AI Agents and David Luan's Background

Introduction to AI Agents and David Luan's Background

03:39David Luan's Journey in AI and Reflections on GPT-5

David Luan's Journey in AI and Reflections on GPT-5

08:04The Maturing AI Landscape and the Focus on Practical Applications

The Maturing AI Landscape and the Focus on Practical Applications

11:32Defining AGI and Amazon's Strategic Vision

Defining AGI and Amazon's Strategic Vision

15:57The Current State and Future Promise of AI Agents

The Current State and Future Promise of AI Agents

20:53Overcoming Limitations of LLMs and Training Agents for Real-World Decision-Making

Overcoming Limitations of LLMs and Training Agents for Real-World Decision-Making

25:35Amazon's Approach to Training AI Agents: Large-Scale Self-Play

Amazon's Approach to Training AI Agents: Large-Scale Self-Play

27:30The Future of AI Agents and the Importance of Product Form Factors

The Future of AI Agents and the Importance of Product Form Factors

29:59Amazon's Internal Collaboration and the Strategic Advantage of Private Data

Amazon's Internal Collaboration and the Strategic Advantage of Private Data

32:51Internal Adoption of Agents within Amazon and the Focus on Reliability

Internal Adoption of Agents within Amazon and the Focus on Reliability

36:26The Timeline for Advanced AI Agents and Amazon's Leapfrog Strategy

The Timeline for Advanced AI Agents and Amazon's Leapfrog Strategy

41:50The Talent Market and the Reverse Acquihire Phenomenon

The Talent Market and the Reverse Acquihire Phenomenon

45:26The Rising Costs of AI and the Importance of Partnering with a Major Player

The Rising Costs of AI and the Importance of Partnering with a Major Player

49:06The Limited Pool of AI Experts and Advice for Navigating the Field

The Limited Pool of AI Experts and Advice for Navigating the Field

54:34Conclusion and Episode Wrap-up

Conclusion and Episode Wrap-up