Enterprise Multi-Agent Systems in Codex

This intensive, hands-on workshop teaches participants to engineer a working, enterprise-pattern multi-agent research system directly in Codex. Using a realistic consulting scenario, teams build a workflow that gathers evidence, validates calculations, challenges assumptions, and produces an executive-ready recommendation supported by traceable evidence.

Modeled on Kearney's internal Research OS benchmark, the workshop covers the complete system lifecycle: workflow auditing, operating contracts, specialist subagent delegation, controlled tool integration, deterministic validation, adversarial review, and reviewable delivery.

Working in approximately 17 teams of three to four from a prepared but intentionally incomplete repository, participants complete and test a Codex-native Research OS workflow. The primary workflow delegates specialized work to Source Research, Benchmark Analysis, Calculation Check, and Red Team subagents defined in .codex/agents/. Agents exchange structured JSON artifacts for sources, claims, calculations, caveats, and open questions instead of relying on unstructured conversational handoffs.

Participants leave with a functional Codex workspace, reusable project assets, a packaged Codex Skill, and a validation pipeline that can be extended to enterprise consulting engagements.