Working with multi-agent systems in production has been tough, especially when trying to maintain strict latency budgets as the number of agents increases. Chaining multiple calls often blows out our request times unexpectedly latency budget constraints in AI