Kiro Agent Workflows & Eval Harnesses

The Goal

I wanted repeatable AI workflows that improved engineering output instead of creating inconsistent one-off results. The focus was not just getting a model to respond, but shaping a workflow that could move from ideation to implementation to verification with better reliability.

What I Built

• Structured workflows: Built a set of AI workflows for Kiro that supported ideation, implementation, verification, AWS documentation lookup, and note-taking with Obsidian integration.
• Repeatable developer loops: Designed the workflows so they could be reused across both software engineering and solutions architecture tasks.
• Evaluation before release: Used promptfoo to test prompts and workflow behavior before publishing changes, specifically looking for contradictions, weak instructions, and brittle behavior.

Results

• Among the top 75 most active users out of 275k+ internal Kiro CLI users.
• Reduced my own administrative overhead by roughly 4 hours per week.
• Created a more reliable personal workflow for using AI as a systems-thinking partner rather than an autocomplete layer.

Why Employers Should Care

This work demonstrates a practical point: productive AI use is not about collecting prompts. It is about designing workflows, constraints, evaluation loops, and operating habits that make the output more useful and less noisy.

That is the kind of thinking teams need if they want AI adoption to improve execution instead of just increase experimentation.