Addy Osmani explores the concept of the 'software factory' — a system of harnessed AI agent loops running at scale to produce code. The post distinguishes between 'light' factories (humans remain in the review loop) and 'dark' factories (fully automated, no human reads the code). Drawing on Dex Horthy's talk and real-world experience running a fully automated code factory for four months, the piece argues that dark factories accumulate 'comprehension debt' silently while tests stay green. The real bottleneck is verification, not generation — human attention is the scarce resource. The solution is keeping humans in the 'outer loop': owning architecture, design decisions, and review gates, while delegating well-bounded, cheaply verifiable inner loops to agents. Good software architecture (types, test seams, clear boundaries) acts as a safety net against agent mistakes and is the key to earning loops their 'lights-out' status.
