You can put crappy products on time or put out polished products when they're ready. It's the Quality/Cost/Time triangle: choose any two, which determine the third. They have a high bar in quality, they've spent hugely already, so time must slip.
I'd much rather a quality product that people rave about than poor reviews and unhappy customers; other businesses seem to disagree.
Having bet and been burned by Elon's timelines previously, I now take his estimates and add a year.
Another way to think about the optimistic time estimates that Elon has doesn't come from project management at all - it comes from chaos theory and dynamic systems when dealing with complex cause and effect.
I claim that project management has a typically unstated and extraordinarily important assumption that the work being managed is sufficiently simple in its cause and effect dynamics that the nature of cause and effect is well understood - at the very least, by experts in the field. For cause and effect to be simple (well understood and widely agreed among both experts and non-experts), whatever it is needs to be something that has been done many times. My mental model for something that falls into this category is the building of a house - clearly many moving parts and a lot of details to get right, and yet it is reasonable to build a project plan before the first shovel of dirt is moved that describes the flow of different trades in and out of a job site, the timing of materials arrival at the job site, and to predict with very high accuracy when the house is complete and ready for move-in. There will be some variance within that plan and the completion, but we expect that variance to be percentage points - not multiples or even the wrong units.
Contrast that with a difficult software engineering project such as autonomous driving. There is nobody, anywhere, that has a fully productionalized and implemented version of that. As a species, we're still figuring out what all needs to be done to complete that "project" for the first time. Any project plan covering that project might be using the wrong unit of measure (decades instead of years, months instead of years - who knows?). We don't expect the same level of predictability in that project plan because its really more like a guidepost that indicates what people were thinking was needed at the point in time when it was created, rather than this particular teams plan for how they're going to do again for the umpteenth time something that people know how to do.
Model X and Model 3 lie somewhere between these two extremes. They are more like the building of a house in that there are many companies around the world design, engineering, and then manufacturing cars every year. To the extent that the Model X and Model 3 are JUST like any other car, we would expect a similar level of project performance from Tesla were Tesla a car company just like any other. To this extent, I currently rate Tesla as poor performing - activities that are routine for other car makers in design and engineering doesn't yet appear to be routine for Tesla.
To the extent that Model X and Model 3 are more like a unique software engineering project that has never been done, I rate Tesla as outstanding.
One measuring stick is to assume that Model X and 3 design and engineering are more like Electric than Gas vehicles - how does Tesla do at that compared to the rest of the industry? Very well when we consider only drivable, buyable vehicles. This latter criteria, in my mind, is the direct equivalent from the Agile Manifesto - "Working software is the best measure of progress". It's not the ONLY measure of progress, but its the best one. Partly because you have something tangible and specific to be learning from.
And this all comes back to learning, at least for me. Tesla is learning how to build things that nobody has yet built (at this quality, quantity, and performance level). Is there equivalent work that Tesla can be better at? I absolutely believe so, just as I believe there are plenty of areas for other car makers to get better at. The question in my mind is (1) is Tesla being shady or misleading with their time estimates and (2) who is going to learn faster and better?
For (1), I don't believe so. Actually, I'd never even thought of the possibility until it was mentioned in this thread. I work on these sorts of complex projects in a different industry - it is immensely frustrating to come up with a project plan of what you think is going to happen as of some point in time, and then later prove to be wrong both in number and unit about what would actually be needed. It's not deceitful - it's the nature of doing something that nobody, anywhere, knows exactly how to do what you're doing. Sidebar - I believe that we're going to see more and more of this kind of work in the Information Economy and we're going to need to get more comfortable with this kind of uncertainty (that's my opinion - doesn't mean I'm right).
I expect its frustrating to Tesla also to think something will be done at point in time X, and then later learn about the unknown unknowns when the X guess was created, and then need to alter the plan.
For (2), this one is easy for me - there is no car company in the world that I see being able to achieve and maintain a faster learning and optimization rate than Tesla. This includes for me, Apple, should they choose to enter the business and until I learn differently about what they are doing. There is no car company that I trust MORE than Tesla to identify things they aren't doing well, and learn from / optimize / improve on those things, and get better. If nothing else, other car companies will need to drop the antiquated "model year" idea because once/year refreshes means their cycle time is at most 1/365th of Tesla's.
In the end, I don't really see the Cost/Time/Quality triangle from project management as being all that relevant to what Tesla is doing. It's too new for anybody, and bringing assumptions from that world and trying to apply them to what Tesla is doing will generate a misleading analysis.
