Google is developing a multi-agent system within Gemini for Enterprise that automates research and idea generation. This system can take a topic and evaluation criteria, produce around 100 ideas in a single session, and evaluate them in a tournament-like format over a continuous 40-minute period. Users then receive a ranked list of ideas, complete with detailed descriptions, performance reports, and summaries for each suggestion. This approach allows for deep exploration of specific options and is designed to offer more substantial output than typical AI tools. Currently, there are three pre-made agents in the preview stage. The “Idea Generation” agent focuses on generating and ranking ideas based on user-defined topics. The “Co-scientist” agent caters specifically to researchers, enabling them to input additional data and receive structured research directions. Both agents utilize a tournament evaluation system, which appears to allocate significant computing resources, a notable feature compared to existing models. This extended runtime allows for thorough problem-solving, aligning with the concept of "Level 3 AI." While Gemini for Enterprise is still under development and not available to the public, these multi-agent workflows and the document-focused agent, which allows users to interact with uploaded PDFs, represent innovative capabilities. The potential of these tools, especially if powered by the upcoming Gemini 3 Pro, could significantly impact organizations and research teams. However, concrete evaluations and benchmarks are necessary to fully understand their effectiveness. Overall, the advancements in Gemini suggest a shift in how AI can assist in structured research and idea generation.