Software factories systems that continuously build, test, and improve software with increasing autonomy are moving from experimentation to infrastructure. But making a coding agent work is different from engineering a reliable factory at scale. Giving every engineer an agent does not create a software factory. As autonomous work scales, organizations must manage shared codebase context, integrate independently produced changes, and enforce engineering standards without restricting developer autonomy. This talk explores how open-weight models and open-source infrastructure enable organizations to build software factories around their own engineering practices. We examine architectural choices behind shared context, verifiable execution, extensible pipelines, and human-agent collaboration. We also explore how an open stack provides greater control over performance, cost, data, model choice, and integration, while systematic verification improves the cost and reliability of software production. Santosh Kumar Radha, CTO of AgentField, shares practical lessons from building software factory infrastructure and creating a shared production capability that improves with every run.