Google DeepMind is using artificial intelligence to accelerate its own research, deploying AI agents alongside scientists to identify gaps in its Gemini models and speed development, according to two senior executives. The approach marks a shift in how the company’s India teams work, using AI to help chart a path toward artificial general intelligence (AGI), which can perform human-like tasks across domains. Demis Hassabis, Google DeepMind’s chief executive officer who shared the 2024 Nobel Prize in Chemistry, has said that AGI could arrive as early as 2029. By assigning routine testing to AI agents, DeepMind hopes to speed up specialised applications in drug discovery and materials science, while strengthening its foothold against OpenAI and Anthropic in India’s developer market. Speaking at the India AI Impact Summit in February, Hassabis proposed an “Einstein Test” for AGI: Training a model exclusively on pre-1911 scientific data to see if it can independently derive general relativity. Manish Gupta, senior director for India and APAC, and Seshu Ajjarapu, senior director of applied AI, told Business Standard in an interview in Bengaluru that teams, including those in India, are using AI agents to improve Gemini by identifying capability gaps, curating data and verifying improvements. Teams in Bengaluru are improving Gemini’s reasoning and multilingual performance. Gupta said automating parts of AI research gives scientists tools for challenges such as disease discovery and new materials, with work done globally, including in India.