Building an agentic software factory? Start with the sensor.

The industry has long struggled with the gap between the high shipping velocity of top engineering teams and the rest. Traditionally, this gap was measured in deploys per week. However, with the rise of "software factories," a largely autonomous system where intent goes in, agents plan, write, test, and ship changes, this gap is rapidly shrinking.

Anthropic recently reported that in May, they merged around 2,500 code changes a day, and by September, that number skyrocketed to over 30,000. It's clear that coding is no longer the bottleneck; instead, it's the developer infrastructure built for human scale that has become the limiting factor.

The key to unlocking this transformation lies in rethinking the delivery lifecycle around a crucial question: how does the factory know a change works before all the others land on top of it? To achieve this, engineers must rethink the delivery lifecycle around a central question: how does the factory know a change works before the rest flood in?

Control engineers emphasize that a controller must outrun its sensor, which means the sensor must be able to keep up with the controller's pace. In the context of software factories, the sensor plays a critical role in determining how much of the system can run autonomously. An agentic software factory comprises many interconnected loops, fed by a work queue and drained through a review gate into production.

Humans own the entire process from above, with intent represented by specifications, tickets, and scenarios, and verification handled by tests, scenario runs, and runtime checks.

A controller that fails to outrun its sensor will fail to improve speed and may even introduce errors by relying on stale information. Therefore, the sensor must accurately reflect the system's behavior to ensure the system's smooth operation.