I'm a huge fan of the changes described in Kevin Carson's The Homebrew Industrial Revolution: A Low-Overhead Manifesto.
I love the fact that technology is enabling things like the re-localization of manufacture and a "small batch" ethos that lets people get something a lot closer to what they want instead of just having to settle for whatever one-size-fits-all model some megacorp produced a billion of.
If there's a weakness there, I think it looks something like this [hat tip -- Judy Morris].
Fisker has built only 239 units of its 2012 Karma hybrid car, and is recalling them all because of a possible fire risk from coolant leaks.
Yes, the advantages of product continuity, economies of scale, etc. are over-rated in some important respects. But producing large quantities of a product that's altered incrementally does lend itself to gathering more data from which problems can be detected and predicted.
Just as a ferinstance, as of 2010, Ford had sold more than 2.3 million Focuses over a 12-year period. Presumably real-world-experience information gathered from each previous year's model (and over the history of its predecessor, the Escort) was used to improve the current one.
When you've only made 239 of a car, and only put 50 of those on the road, there's a lot less specific data to generalize and improve from.
And what if the actual build of the car is done not by "repeat the same action over and over" assembly line workers, but by the actual customers; and not at one facility, but at one of a number of "micro-factories," as with the ultra-cool Local Motors Rally Fighter? It seems that would make it a lot harder to reach a determination along the lines of "ah, that's what's causing those breakdowns -- we should change the design to call for x pounds, instead of y pounds, of torque on that bolt." Because you really have no way of knowing if your customer who built his car from your kit actually put x pounds of torque on the bolt, do you?
Then again, if you only produce 239 cars, I guess you don't have to worry about recalling 3.8 million at one whack, do you? So if problems are more likely to go undetected/unpredicted in early design/testing, they're also less widespread and easier to correct when you do detect them.
And the smaller the batch and/or more bespoke the final product, the more it's a case of people getting what they actually want instead of what some bureaucratic suitie in Detroit decided they should have. Which, I think, goes a long way toward balancing out increased risk of undetected/unpredicted flaws. Especially since the Big Guys haven't actually eliminated that risk, and in at least some cases seem to have just factored it in as a risk worth taking versus the bottom line.