Enterprise Multi-Agent Decision Systems with Claude

This intensive, hands-on workshop teaches participants to engineer a working, enterprise-pattern multi-agent decision system directly with Claude. Using a realistic consulting/procurement scenario, teams build a workflow that gathers evidence, validates calculations, challenges assumptions, controls cost through model routing, and produces an executive-ready recommendation supported by traceable evidence and a clear path to production consumption.

The session walks participants through the complete lifecycle: from workflow audit and operating contracts to specialist subagent delegation, controlled tool integration, cost-aware model routing, deterministic validation, adversarial review, production hand-off, and reviewable delivery.

Working in teams of 3 to 4 from a prepared but intentionally incomplete repository, participants complete and test a Claude-native decision system. Project instructions are configured so the primary Claude workflow delegates specialized work to four subagents (Contract Analysis, Policy Compliance, Alternative Recommendation, and Red Team). The agents exchange structured JSON artifacts for sources, claims, calculations, caveats, and open questions rather than relying on unstructured conversational handoffs.

Participants leave with a fully functional Claude workspace, reusable project assets, a packaged skill, a working validation pipeline, basic cost-routing rules, and a clear production hand-off pattern that can be extended to real enterprise decision and repetitive-task workflows.