Open-Source AI Software Factory — Airflow × Crabbox

​Click here for more information and to register: https://github.com/zozo123/ariflow-swfactory/issues/2122

​Coding agents can generate patches quickly. The harder problem is operating them as a reliable engineering system: bounded execution, durable work identity, retries, human approval gates, reproducible verification, credential boundaries, retained evidence, and deterministic convergence before anything reaches main.

​In this hands-on workshop we build a small, inspectable open-source AI software factory. A GitHub issue becomes a specification and plan; Apache Airflow owns the durable lifecycle, mapped work, retries, waits, and approval gates; disposable execution performs build and test work; Crabbox demonstrates remote runner leasing, worktree synchronization, command execution, streamed output, and retained evidence; then deterministic verification produces one reviewable pull request while humans retain final merge authority.

​The core design rule is: Parallelism is cheap. Authority is singular.

​Airflow Software Factory: /zozo123/ariflow-swfactory

​Crabbox: https://crabbox.sh/

​Apache Airflow: https://airflow.apache.org/

​Independent open-source community workshop using Apache Airflow, Crabbox, and the Airflow Software Factory project.