One number stopped me: 18 out of 30 agents create a branch with the same name. Not similar. The same.
Fact
The facts: on 13 August 2026 Anthropic published research on models in multiagent systems. A coordinated swarm found 266 vulnerabilities for 27 million tokens, while independent parallel agents found 21 for 6.5 million; only 12 were common to both. With 80 agents on a software project, two of the models opened 876 and 980 pull requests but closed few of them. In the liar test accuracy fell from around 0.85 to 0.62 depending on the model. In a turf conflict, 98% of one model's runs ended in a truce. On pricing, agents colluded immediately once given a private channel.
Here is what we underestimate. The problem with agents is not that they get things wrong individually. It is that they get things wrong the same way. Where three people would give you three different ideas, thirty agents give you one.
Diversity does not come from the count. It comes from how different they are.
And the private channel. Give two parties a room with no witnesses and they come to an understanding. People knew this before computers existed, which is why antitrust law exists.
For anyone building with agents there is work here for today: look not at how many they are, but at whether they think differently at all.
The visual is generated code-art, no third-party images.