How third moment research manages 10,000 autonomous AI agents for extended investigations

Third Moment Research is leveraging advanced AI models like GPT-6 Astra, GPT-6 Sol, and GPT-6.1 Sol to run 10,000 autonomous agents simultaneously for investigations lasting up to 36-48 hours. These agents are designed to improve with use, focusing on specific tasks and adapting to workflows unique to each company. The system uses synthetic scenarios to identify weaknesses and enhance agent capabilities continuously. In quantitative finance, agents analyze trading strategies and optimize backtesting processes, reducing execution times from hours to minutes. Codex, integrated with GPT-5.6, identified and resolved bottlenecks, improving efficiency and reducing token usage. Adam, a researcher at Third Moment, highlights how agents now autonomously design follow-up experiments, leading to unexpected insights, such as the impact of non-displayed trades on market volume. The company also employs Daybreak for cybersecurity testing, ensuring agents cannot breach their research environments. Third Moment aims to scale further, planning to deploy hundreds of thousands of collaborating subagents to autonomously improve AI systems as new challenges arise. Adam envisions a future where humans guide AI research more strategically, optimizing how agents learn and operate.