Looks modern too. But here’s the problem: most teams are stacking subagents on top of weak context, and that breaks the whole system. If the model gets the wrong tools, stale memory, noisy history, or missing repo facts, your shiny agent setup turns into expensive conf