OpenProse Team Demonstrates Declarative Contracts for Multi-Agent Workflows: Define Intent, Not Steps

For developers building complex AI agent systems, managing the collaboration logic between multiple agents is a core challenge. In this episode, Dan Barrett, Raymond Weitekamp, and Jose Montes de Oca from the OpenProse team propose a new approach: stop telling agents how to do things, and instead declare what you want.

The team demonstrated how to write multi-agent workflows as declarative contracts using plain Markdown. The advantage of this format lies in its universality; any execution environment supporting the standard (such as Claude Code, Codex, and others) can compile and run it. This marks a shift from tedious prompt engineering to higher-level intent declaration.

The video includes several specific application demos. First are event-driven jobs triggered by webhooks and pull requests, showing how contracts respond to external changes. Next is a PR reviewer styled as a "grumpy senior," illustrating the ability to define specific behavioral patterns through contracts. Finally, the team demonstrated an end-to-end bank model documentation workflow, highlighting the potential of this method in handling complex business logic.

Beyond basic usage, the discussion covers advanced topics such as composing contracts, pinning successful runs into reusable workflows, and comparing the OpenProse approach with the DSPy framework. For technical professionals looking to optimize agent orchestration efficiency, this content offers practical directions worth exploring.