Inside a Software Factory: How AI Agents Move Through the Software Lifecycle
In a column for O’Reilly Media’s Radar, software engineer Paul Iusztin argues that development teams should focus on building software factories rather than obsessing over trendy engineering labels.
After more than seven years of pre-AI coding experience, Iusztin built a factory named Squid to ship small and medium projects for Decoding AI with minimal human intervention. The initial version proved so overbuilt that its creator stopped using it.
Iusztin writes that recent industry buzzwords like loop engineering and graph engineering sound more like marketing talk than solutions to real problems. He says that Boris Cherny, who leads Claude Code at Anthropic, stated, “My job is to write loops.”
Instead of overexplaining these concepts, engineering teams should focus on the software factory framework. Tereza Tížková of Factory.ai defined the model at the AI Engineer World’s Fair 2026 as “the whole loop, the whole lifecycle of developing software with autonomy,” according to news.lavx.hu.
Everythingblog.in noted that the operational framework relies on an eight-stage architecture categorized into planning, building, and self-improving loops. Raw work enters the system, and shipped software comes out the other end.
The eight stages span triage and intake, brainstorming, planning, implementing, review, review-CI, release, and monitor or incident response. Addy Osmani frames the stack by stating that “the loop is the atom” and a factory is “an org chart made of loops,” while Warp CEO Zach Lloyd predicts that “software engineering will become factory engineering.”
Paul Iusztin developed the Squid software factory based on his professional background as a software engineer with more than seven years of pre-AI coding experience. This extensive technical foundation informed his approach to integrating coding agents into the development lifecycle, specifically aimed at shipping small and medium projects for Decoding AI with minimal human intervention.
Driven by the necessity to bridge the gap between abstract “vibe coding” and the practical generation of maintainable software, Iusztin launched the Squid project three months prior to his 2026 O’Reilly Media column. His experience led him to prioritize actionable outcomes within the factory framework, despite his initial struggle with overbuilding the system’s early iterations.
Humans retain control over high-leverage strategic decisions during brainstorming and planning gates. News.lavx.hu reported that OpenAI took autonomous execution to an extreme by generating roughly 1 million lines of code and 1,500 merged pull requests over five months with zero hand-written lines, following the principle that humans steer while agents execute.