The most insidious failure mode in agentic engineering isn't buggy code. It's agents building fundamentally misaligned features, products, and systems. Solving that is a bigger and more interesting problem than reducing defects.
This is why I think the "software factory" is the wrong metaphor for AI engineering. The challenge is less how to stamp out the same door panel every time with Six Sigma quality than how to evolve a system in line with our vision, values, and architecture. In that sense, the process is closer to Andy Warhol's Factory than Ford's car factory. Both are focused on throughput, but Warhol was more concerned with ensuring all work aligned with a single creative vision.
The hardest problem for a business is still creating a vision and keeping an entire team of humans—and now humans and agents (and humans with agents)—building toward it, from the system architecture down to the individual lines of code. As I learned long before agents existed, achieving this is much more like building a startup than assembling a car.
Too much of the industry treats software as a problem to be optimized and solved. That may be true for code writing and testing, but the better metaphor is staring us in the face: It's a software company, not a software factory.
A factory is one piece of a larger organization, where layers of interdependent systems interact and move at different speeds. But a company is, and always has been, a collection of agents brought together to collectively build something. Understanding the motivations of those agents is what it takes to build a successful "software factory" and, conveniently, exactly the job of a software company CEO.
