Claude is performing 3.8x poorer compared to last week based on my claude code token analysis

3 issues came to light: Each sub agent is re-reading more from the parent agent Each subagent is taking 2x steps to finish the job Claude is spinning more sub agents per task (this could subjective per task type) Hypoth…