Test-time scaling has two axes: running agents over longer timeframes (depth), and running a larger number of agents (breadth).

Everybody knows about the first axis, but the second one is just as important when solving hard problems that require broad search.