I'd agree lots of software is like engineering in that it is further tilted towards the practitioners and marketplaces and their limitations. It has more focus on cost-effective repeatability and risk management. But, traditional engineering is more rooted in some physical reality. I.e. civil engineering has basis in physics and materials science, because those drive failure modes and design criteria. The more social aspects---say evacuation routes for fire safety---developed within an existing practice that already been formalized to manage risks of structural failure etc.

But, lots of software is far detached from such a physical realm. It is more like legislation or contract law, or the arts. It's all about what people want to say or hear and display to each other. Massive influence from the cognitive and social context. It may be colored by technique and available materials or resources, but thanks to Moore's Law and friends, these practical limits have less and less influence on the shape of the outcome. Significant failure modes can be rooted in the vagaries of psychology, social sciences, or even political sciences.

However, software diverges from law or arts when we attempt to scale it up. We want to reach speeds that cannot be supervised and quality controlled by a human operator, or volumes of support tasks that cannot be managed by an affordable human workforce, or impactful decision-making and effects that cannot be adjudicated by a higher court. You need something like engineering here to manage risk, but there is no physical basis for building codes etc. Not only the new software products, but even the elements of risk are social constructs.

But I think the metaphors are broken as one pushes too hard "up the stack", into AI replacing knowledge workers and analysts. Our current best social systems completely depend on human adjudicators in hierarchical layers of oversight. Courts that can comprehend the big picture and resolve differences when our other abstract social activities go wrong. I am nervous about a future where tech behaviors become too complex for such oversight. Or worse, where naive practitioners think they can solve it with another layer of unproven tech. Instead of turtles all the way down, it becomes some faith-based electronic oracles all the way up?