Abstract

An AI agent harness wraps around a Kanban board and builds software stories into code changes. The core is deterministic: code decides the next stage through a fixed pipeline, and AI models only reason inside a step, with their output checked by code. A scheduler starts short-lived workers. Each run advances one story by one step, then exits, after claiming the story atomically from a ready queue. The harness stops and hands work back when requirements contradict, when a plan can't be completed, or when checks fail. A code-only gate approves builds into a Staged column, and a human merges.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Share

COinS