Third Moment Research is demonstrating a major step forward in scalable AI agent systems: running 10,000 agents for more than 36 hours. That kind of sustained operation suggests AI agents are becoming more capable of handling long, complex workflows rather than only short, one-off prompts.
The lab is using a combination of OpenAI technologies, including GPT-6 Astra and GPT-6.1 Sol, Codex for coding, and Daybreak for cybersecurity. Together, these tools allow agents to stay focused over extended periods, coordinate work, and improve through continued use.
Why this matters
Persistent agent swarms could unlock new productivity in research, software development, and security operations. Instead of waiting for humans to break large problems into tiny steps, teams may be able to delegate broader goals to AI systems that can work for days, monitor progress, and adapt as they go.
- Scale: 10,000 agents running in parallel points to powerful orchestration capabilities.
- Endurance: 36+ hour runs show progress toward reliable long-duration AI work.
- Practical impact: Coding and cybersecurity are early areas where persistent agents could deliver real value.
While this is still a research milestone, it is an encouraging signal for the next generation of AI infrastructure: systems that do not just answer questions, but persistently collaborate on meaningful work.