Skip to content

Design Patterns for AI Workflows

Design patterns for AI workflows are reusable templates and architectural approaches for structuring how researchers interact with AI tools — analogous to software design patterns.

Context & Background

As researchers gain experience with AI tools, common patterns emerge for structuring effective interactions:

  • Prompt-Plan-Review-Revise (PPRR): Iterative cycle of planning and refinement
  • DAAF (Data Analyst Augmentation Framework): Structured framework for AI-assisted data analysis
  • Compilation and review: AI compiles information, human reviews and curates
  • Skill-based architecture: Reusable AI instructions packaged as "skills"
  • Pipeline pattern: Sequential processing stages with defined inputs and outputs
  • Fan-out/gather: Parallel processing of independent tasks followed by synthesis

Practical Implications

  • Learn the common patterns: Understanding established patterns saves you from reinventing them
  • Match pattern to task: Different research tasks call for different interaction patterns
  • Document your patterns: Write down workflows that work well so you and others can reuse them
  • Evolve patterns over time: Refine your approaches based on what works