Writing complex multi-agent workflows can often feel like an intricate dance where you must carefully orchestrate each step. Enter OpenProse, a groundbreaking approach that shifts the paradigm from “how” to “what,” allowing developers and AI engineers to declare their intentions through plain markdown as declarative contracts. Instead of outlining the detailed steps to achieve a goal, OpenProse lets you specify your desired outcome, and the system works out the specifics, making your workflows both efficient and elegant.
At the recent discussion featuring Dan Barrett, Raymond Weitekamp, and Jose Montes de Oca from the OpenProse team, they showcased this novel method of interaction with multi-agent systems. The core idea is simple yet revolutionary: say what you want, not how to get it. This philosophy not only simplifies the creation of multi-agent workflows but also opens them up to being compiled and executed by any harness, such as Claude Code or Codex.
Based on content from AI Engineer
Understanding the OpenProse Approach
OpenProse operates on a two-tier concept: it’s all just markdown, and it’s event-driven, not conversation-based. By isolating these two elements, OpenProse offers versatility and accessibility. Anyone familiar with markdown can read and understand these workflow contracts, which are designed to be human-readable and modifiable in logical English. This makes them accessible to professionals even if they do not come from a coding background.
The team’s demos illustrated how these declarative contracts can form the backbone of robust systems—from workflows triggered by webhooks and pull requests to event-driven job executions. These job executions are driven by contracts and specific triggers like cron jobs or external events, such as GitHub actions. The system enables you to express your intent, compose your contracts, and pin successful runs into reusable workflows, saving you the hassle of starting from scratch each time.
Demo Highlights: Real-World Applications
One notable example shared by the OpenProse team involved a “grumpy senior” PR reviewer. In this scenario, the system autonomously reviewed pull requests adversarially, reducing bias and offering a more objective assessment. Another exciting demonstration was the end-to-end bank model documentation workflow, showcasing OpenProse’s capability to manage complex, regulated processes seamlessly. The flexibility and clarity of the markdown-based contracts make such tasks straightforward and maintainable.
OpenProse vs. DSPy
During the comparative exploration of OpenProse against DSPy, it was evident that OpenProse’s event-driven methodology offers a clear differentiation by focusing on specifying results rather than procedures. It allows AI engineers to couple the event-driven processes with declarative specifications, thereby enhancing efficiency and promoting reusable code.
Looking Forward: OpenProse and The Community
In addition to its technical prowess, OpenProse emphasizes community engagement and collaborative improvement. The open-source nature means it’s not only easy to adopt but also continuously evolving through community input.
As AI becomes more integrated into various aspects of industry and daily life, solutions like OpenProse that democratize access and simplify complex workflows will be pivotal. As highlighted by Dan Barrett and his team, understanding and incidentally leveraging these capacities can indeed reshape the future of AI engineering.
For those intrigued by this transformative technology, the upcoming second annual AIE CODE summit offers an ideal opportunity to dive deeper into AI engineering innovations, with applications now live.
