Factories are fine, but if you walk into a factory you're going to be finding a lot of people working on the factory itself. Process tweaks, new machines, maintenance, etc.
In the software "feature factory", people have mostly forgotten about that, usually because someone does it "in their spare time". I guarantee you that no factory worker lubricates the machines off the clock. I am not sure why we should treat software any differently.
A factory can produce almost anything at scale, but how do you know it is worthwhile and providing the best possible value? In a "bad" factory, the "factory workers" are like cogs in a machine and have no influence over what is built. There is someone ordering to build this or that and they produce it. Feature factory is in this terminology should be seen as opposed an alternative process, where the "factory workers" themselves decide what to produce and how, and as long as the measured value stacks up, it will be more efficient and derive more shareholder value than the "feature factory".
Factories are efficient because they make a bunch of things in the same way. The feature factory isn't a factory in that sense. It's a giant craft shop in which every job is unique and no real scaling is occurring.
The problem is that "shipping stuff" is not what companies are supposed to be solving. They're supposed to be providing _value_ to customers and investors.
A "feature factory" model is bad because it masquerades as progress.
"Value" is harder to define, and often can't be measured _simply_ with metrics - you need metrics for insights, but most metrics are very much trailing indicators. Also there are lots of silly metrics like "tickets closed" that are easy, and naturally companies gravitate towards anything easy as the number of people rises. And factories love metrics.
Which is great, so long as they're not just creating debt for themselves by e.g. creating broken products that have to be recalled to be fixed by the factory.
when it comes to software you do not want to produce code at scale. the fewer lines of code the better really. it'd be like measuring the quality of your genome by counting the rungs in your dna.
