A new Stanford study found that when two AI coding agents collaborate on a task, they perform nearly 50% worse than one agent working alone. The bottleneck isn't what you'd expect:

→ View original post on X — @stanfordhai