Production Agent Engineering

This is the production tier: not toy agents, but swarms that survive real load. You'll gate quality with generator-critic loops that evaluate, refine, and refuse to ship below your bar; orchestrate long-running task agents by goal — polling for stalls, capping turns, and killing the infinite loops that wreck real systems. You'll trace every hop of a multi-agent run (who did what, where it broke, what it cost), guardrail the swarm by validating inputs and outputs, fencing tool access, and keeping a human in the loop — and finally build an agent that builds agents: a reusable library that loads MCP tools on demand. You leave with the observability, evaluation, and reliability craft that separates a demo from something you'd actually run — and control of every agent in it.