I’ve been using a set of reusable workflow patterns when working with AI, and I often apply the same way of thinking to coding and non-coding tasks.
What interests me most is not any pattern in isolation. It is how the patterns can be combined and nested depending on the problem.
Plan → Execute → Verify
Break down the work → execute the steps → verify the result.
Prompt Chaining / Pipeline
Arrange dependent steps so the output of one becomes the input for the next.
Fan-out / Fan-in
Split a problem into independent perspectives or workstreams → process them separately, sometimes in parallel → synthesize the results.
Red Team / Blue Team
Probe a proposed result adversarially → address the weaknesses → retest it.
Map-Reduce-inspired decomposition
Split a large problem into smaller pieces → process them → aggregate the results.
Combining the patterns
The point is not to use every pattern in every workflow. The structure should follow the problem: chain dependent steps, fan out when multiple perspectives help, decompose large inputs, and challenge results when assumptions need testing.
One possible combination is:
Plan → Fan-out → Execute → Fan-in → Verify → Challenge → Iterate
Within each stage, I can also use another pattern, such as Prompt Chaining or Map-Reduce-inspired decomposition.
What comes next
This is the starting point for a series. In the next few posts, I’ll show how I use these patterns, how I choose between them, how I combine them, and how I apply the same ideas beyond coding.