In fact, the multi-agent system built by our

I was quite surprised to see the new agent framework Matrix recently released by Meta.

In fact, the multi-agent system built by our Bionexus was also using a method in which the central orchestra approves and adjusts all requests, but this structural limitation was always a concern.

However, Matrix, which was announced by Meta, completely dismantles this central power, turns the task instruction into a message object, and moves directly between agents, and each agent immediately processes the task that arrived and passes it to the next agent. Bottlenecks disappear when managers disappear, and throughput increases tremendously.

Good bio-data is expensive, scarce, and regulated. Therefore, it is now believed that AI’s ability to generate data or hypotheses that it reads, connects, and creates on its own is essential. Then, it is impossible to generate knowledge of this scale unless a large number of agents work at the same time.

The direction Matrix shows is clear. In the future, knowledge will be created in a distributed ecosystem where dozens or hundreds of autonomous agents work together to constantly create and circulate, not in central control.

If we introduce this distributed collaborative structure to the Nexus Co-Scientist multi-agent system we are creating, we are confident that AI-based experimental design, hypothesis generation, and knowledge expansion rates will increase even steeper than they are now.

I’ll pass it on to my colleagues at work and implement it quickly. A really fun world is opening up. ^^*

출처: https://arxiv.org/abs/2511.21686