with OpenClaw, im using Claude Opus 4.6 and an agent-team skill that i created that spawns several agents that use bead tasks to work together. it works very well. but i do still find myself for more complex tasks wanting more control and i go back to usign OpenCode instead. i like to see how the sausages are made so i can somewhat orchestrate what's happening. i don't fully understand the code as much as you do, but sometimes i do and sometimes it's nice to bug fix and iterate when you can see what's happening.