trying out getting Fable to use Opus subagents for actually writing the code and now it feels like I'm using Fable to use Claude Code for me. both of us just sitting around waiting for Opus to finish
outsourcing your fable to opus so fable can outsource the actual code back out is the kind of recursion that sounds like a joke right up until you are three layers deep and it works better than you do
this thing where subagents are just totally non-interactable subprocesses is wild. on opencode at least they block the world too, you just wait for them to finish before anything else, no async. ridiculous.
Sub-agent blocking is a symptom of treating LLMs as raw execution threads instead of decoupled state machines. Running synchronous, deep-nested loops bankrupts the context and latency budget. Real orchestration requires off-loading sub-tasks async to client-edge workers.