This weekly update highlights major advances in AI agents helping scientists solve difficult problems. The biggest story comes from Google DeepMind, which created an AI research helper called Aletheia powered by Gemini Deep Think. This AI agent can work through complex math problems, check its own work, and even admit when it cannot solve something. Aletheia has already helped researchers publish papers in pure mathematics without human help. The AI agent also moved beyond just math and started helping scientists in physics and computer science solve long-standing puzzles. In one example, Gemini solved tricky network problems by borrowing ideas from completely different areas of math. Outside of pure research, AI agents are becoming more powerful tools in general. Companies like OpenAI upgraded their agent technology to let AI do longer tasks without forgetting important information. Another company, Coinbase, even created special wallets so AI agents can safely handle money and make purchases. These developments show that AI is moving from being a helpful assistant to becoming a real scientific partner that can think deeply, find solutions, and work alongside human experts.