
In this episode of Vanishing Gradients, Hugo Mount Anderson interviews Ravin Kumar from Google DeepMind about the evolution of AI models and agentic systems, particularly focusing on Gemini 3. They discuss the shift from simple chatbots to complex agents capable of self-correction and multi-step tasks, emphasizing the implications for product builders who are constantly rewriting agent harnesses to keep up with model improvements. The conversation covers the "two cultures" of AI agents—deterministic workflows versus autonomous systems—and delves into the product mindset at Google, including the development of Notebook LLM. They also explore the technical aspects of context engineering, the importance of robust evaluation, and the competitive advantages of well-built agent harnesses, highlighting the rapid changes in software development due to advancements in AI.
Sign in to continue reading, translating and more.
Continue