Built an n8n workflow using a two-stage AI agent architecture a "Think" agent and an "Act" agent running in sequence, each powered by OpenRouter with structured output parsers enforcing clean JSON at every step. A configuration node sets the full context upfront, the Think agent reasons through the task using memory and tool access, its output gets parsed and passed to the Act agent which executes based on that reasoning, and the final structured result is returned ready for the next step in any connected workflow.