The best n8n fix I found this month was boring: lower your agent concurrency settings before touching the prompt

I thought GPT was getting worse, but the real issue was too many n8n workflow runs hitting the same LLM path at once. Setting `N8N_CONCURRENCY_PRODUCT