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

By ai.engineer

Categories: AI, Tools

Summary

Stop telling AI agents HOW to do things—tell them WHAT you want instead. OpenProse introduces a declarative markdown framework that lets harnesses autonomously figure out execution, reducing costs and enabling cheaper models to solve complex problems by composing reusable contracts.

Key Takeaways

  1. Shift from imperative to declarative prompting: Write what you want (outcome, goal, end state) rather than step-by-step instructions. This lets more intelligent harnesses optimize execution paths and run on cheaper models.
  2. Prompting as a skill has evolved through 3 distinct phases: token completion patterns (2021), context engineering with RAG/tool stuffing (2022-23), and now contract composition with self-modifying harness topology (2024+).
  3. Use composition over inheritance for prompt architecture. Build reusable declarative contracts that elegantly compose across teams/systems, enabling local complexity management and reducing redundancy in multi-part AI systems.
  4. OpenProse is a markdown-based programming language for agents. All syntax is plain text/markdown, making it accessible while providing formal contract semantics for naming intent that any harness can self-shape and execute.
  5. The core insight: once you can name what you want, any harness can self-modify, fan out, and loop autonomously. Best practices center on declarative intent naming—letting framework-level intelligence handle HOW, not WHO.

Related topics

Transcript Excerpt

My name is Dan. I am the creator of Open Pros. This here is Ray. I'm joined by Raymond and uh Jose, two fabulous uh teammates, co-workers, uh just phenomenal uh creators and engineers all around. I'll introduce them a little bit in a bit. Uh just a little bit of background about what we're what we're about to show you here. Uh 5 years ago in April of 2021, I was named the uh best prompter in the Aluther AI uh back rooms Discord, which uh was a skill at the time that I thought would be tremendously valuable for many years afterward. And actually it lasted I think around like two or three years maybe was the lifetime of of that skill. Uh and I say that because a lot obviously as you all know has changed in the last 5 years. Uh but one thing has not changed and I've been to a lot of uh worksh…

More from ai.engineer