On a built multi-agent AI systems in production. My architecture has evolved over the past several weeks. Claude will remain my “Second Brain” for long-form reasoning, documentation, knowledge management, and strategic planning. I don’t intend to replace that role. Where I’m looking for guidance is the frontier model layer that sits behind multiple autonomous Hermes agents. These agents will perform infrastructure management, security monitoring, research, reporting, trading support, and tool/API orchestration. Most interactions will be API-driven rather than conversational. The infrastructure is built around Apple Silicon (Mac Studio + Mac mini cluster), with local models handling inexpensive first-pass tasks. Only work that genuinely benefits from frontier-level reasoning would be escalated to a cloud model. My interested in real-world operational experience, cost optimization, and architectural decisions from teams or individuals running autonomous agent ecosystems. Appreciate in advance any insights, examples, or lessons learned.