By the end of this session you'll have a production-ready architecture for a multi-agent platform - the diagram *you* drew, not the one I showed - ready to start implementing the next morning.
Most "agent architecture" talks stay abstract. This one doesn't. We take one concrete enterprise scenario and design it together, decision by decision: agent topology and boundaries, model strategy (big model vs routing to small/local), state and memory, guardrails and the MCP attack surface, evaluation, and the CI/CD + observability that keeps it alive in production. At each gate I'll show what we chose at Payoneer and what we ripped out - then you decide for your case and fill in your diagram.

You'll walk away with a completed reference architecture, a template to implement from, and a decision framework you can reuse for the next platform. Bring a laptop; leave with a blueprint.