Stop Telling Agents How — Dan Barrett, Raymond Weitekamp & Jose Montes de Oca, OpenProse

Why watch

OpenProse presents a declarative approach to multi-agent workflow design where developers specify intent (what they want) rather than imperative instructions (how to get it). The session demonstrates writing AI agent contracts in plain markdown that can be compiled by harnesses like Claude Code and Codex, covering event-driven jobs triggered by webhooks/PRs, a grumpy senior PR reviewer, and a bank model documentation workflow. This approach contrasts with DSPy by focusing on human-readable contracts rather than optimization-focused programming. Ideal for developers building AI-powered automation systems who want more maintainable agent configurations.