Stop Babysitting Agents, Start Authoring Outcomes
Summary
OpenProse is an open-source natural-language programming system designed to transform successful AI agent sessions into reusable, reviewable programs. It addresses the "babysitting" problem of current AI agent workflows by allowing developers to describe multi-step work in logical English, which is then converted into a ".prose.md" program. This program is "compiled" and executed by coding agents such as Claude Code or Codex, rather than an external framework. OpenProse aims to make agent work versioned, inspectable, and repeatable, preventing valuable workflows from being lost in chat history. It functions as an agent skill, enabling agents to interpret logical English contracts that define requirements, outcomes, services, and execution steps. The system also includes "session-to-prose" to extract workflows from existing JSONL session logs and supports explicit declaration of agent skills as dependencies.
Key takeaway
For AI Engineers struggling with unreliable or irreproducible agent workflows, you should adopt OpenProse to formalize your multi-step AI tasks. This system allows you to convert successful Claude Code or Codex sessions into versioned ".prose.md" programs, enhancing trust and repeatability. By explicitly defining requirements and outcomes in logical English, you can move beyond "babysitting" agents, ensuring consistent results and inspectable audit trails for your critical applications.
Key insights
OpenProse transforms ephemeral AI agent chat sessions into reusable, reviewable, and versioned programs using logical English contracts.
Principles
- Agent workflows need explicit contracts.
- Isolate sub-agent sessions for clarity.
- Determinism comes from real tools.
Method
Describe multi-step agent work in logical English, use "prose write" to generate a ".prose.md" file, then "prose run" with a coding agent.
In practice
- Convert successful agent sessions to ".prose.md" programs.
- Declare agent skills as explicit dependencies.
- Use "npx skills add openprose/prose" to get started.
Topics
- AI Agents
- Workflow Orchestration
- Natural Language Programming
- OpenProse
- Claude Code
- Codex
- Version Control
Code references
Best for: AI Architect, AI Engineer, MLOps Engineer, Machine Learning Engineer
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by Turing Post.