Love the bottleneck framing here - most automation discussions feel like optimizing random tasks rather than actually thinking systemically. I've been running a local agent setup for somecode review stuff and the token budgeting thing is real, especially when context balloons. The shift from CEO model to intern-as-router makes way more practical sense than throwing Opus at every tiny task.
Love the bottleneck framing here - most automation discussions feel like optimizing random tasks rather than actually thinking systemically. I've been running a local agent setup for somecode review stuff and the token budgeting thing is real, especially when context balloons. The shift from CEO model to intern-as-router makes way more practical sense than throwing Opus at every tiny task.
Yes there's so much that LLMs can be used for that without bottleneck analysis it is very hard to prioritize!