Token-Length Latency

Testing subagents in a real-world scenario revealed that pushing up to 35,000 tokens in a single context can lead to unresponsive behavior and delays.