Carlos Ospina Proposes 'Agentic Recipes' for AI-Assisted Development

Reflecting on the development of AI-assisted tooling for Drupal projects, Carlos Ospina has proposed a concept called "agentic recipes," which he describes as reusable orchestration plans for implementing project-specific capabilities through AI workflows.

In the post, Ospina traces the evolution of his tooling from workflow stages such as scope definition, research, implementation and review to collections of atomic guides and project-level playbooks. He argues that these approaches work well for individual rules and decisions but become less effective when AI agents must coordinate multiple project-specific choices to deliver a complete feature.

To address that gap, Ospina proposes agentic recipes as an intermediate layer that defines a capability, its prerequisites, expected outputs and the sequence of decisions needed to achieve it. Although the examples focus on Drupal development, he suggests the pattern could apply more broadly to software projects where repeatable workflows depend on project-specific implementation choices. The article is presented as an exploration of whether other developers working with AI agents encounter similar orchestration challenges.