The concept of software factories is making a comeback, driven by artificial intelligence models that can generate and manage code at unprecedented speed, according to a recent analysis published by ZDNET. The report examines how leading AI companies are building similar infrastructure to transform prototype applications into production-ready software through automated, repeatable processes. This revival addresses a fundamental challenge: organizations now face daily software releases and updates that human teams can't validate fast enough.
Major AI companies including Anthropic, Cognition, Cursor, Factory, Google, Github, OpenAI, and Ramp have independently developed software factories that share the same basic architecture, the report finds. Each system includes six core components: a queue where work arrives as issues rather than prompts, a control plane providing a durable surface beyond individual laptops, disposable sandboxes created for each task, pull requests as the output unit retrieved by humans, event streams that monitor every action, and durable memory where only file-written data survives. These companies represent what the report describes as the cutting edge of using agents to automate software development lifecycles. The pattern has emerged over roughly the past 18 months as these organizations built what the report characterizes as "the same machine" while working independently.
According to Moritz Plassnig, CEO of CloudBees, coding previously "created the bottlenecks for a lot of organizations" because hiring engineers was difficult and expensive. Foundation AI models and associated agents have changed that dynamic, making organizations "less constrained on the coding side," the report states. Developer roles are shifting from writing code to "exercising judgment around what gets built and shipped," Plassnig explained. The report also highlights a critical limitation: while code generation has become easy and cheap, verification remains "a major engineering problem." As one engineer quoted in the analysis warns, building a software factory doesn't excuse verification or allow degraded codebase quality.
The report explains that software factories address the challenge of managing hundreds of thousands of lines of code alongside releases that now occur daily or even more frequently at many organizations. Plassnig illustrated the problem with automation limits: when changes happen every few minutes rather than periodically, human monitoring of alerts, error rates, and resource usage breaks down. The factory model enables someone with a software idea—such as a customer support worker responding to feedback—to submit a prototype that gets created, reviewed by product and engineering teams, and refined through the factory process. This approach mirrors manufacturing, where a car prototype moves from experimental design to scaled production. The concept isn't new; Microsoft discussed software factories in 2008, but the idea faded until AI's coding capabilities revived it. Organizations can now act on ideas quickly while maintaining quality standards, but only if they solve the verification challenge that has become the real blocker for companies at the frontier. The trade-off is clear: automation can eliminate coding bottlenecks, but quality assurance must evolve just as rapidly or the entire system risks producing faster failures instead of faster innovation.