I have AP1. I consider it spectacular value for the $2500 it cost during ordering time. It works in conditions where I never thought it could. In general, I assumed AP was a gimmick - a cool adaptive cruise control with some laneholding ability that required frequent intervention, at best. In reality, it's far better than that.
It handles all kinds of freeway traffic, and even remained engaged for over an hour through near whiteout rainfall conditions when lanes were barely discernible to the human eye without complete focus. I had my hand on the wheel and was surprised it could handle such conditions where it was a significant effort even for *me* to see the road. The wiper rate was up and down, and yet the camera didn't seem to be perturbed by the wiper cutting its view. That was tremendous.
There are certain conditions that fail AP, and I've driven enough to know when it's likely to get uncertain, e.g. cresting a hill or bright sunshine hitting windshield.
The greatest value AP1 brings is the extent to which it has kept improving. It used to be much more unsteady when there was an exit to my right. Now it avoids the issue much more effectively. And that's one of many things.
AP2 is still evolving. Will it one day be worth $5K (or 8K for FSD) ? Probably. But today, AP1 is worth more than I paid for it and AP2 is worth less than what I'd pay for it today, for me. My concern with AP2 is that it takes upon tasks that entail regulatory oversight, e.g. intervention into and out of traffic, as opposed to staying within a traffic lane. AP1 does this only to a very limited extent, i.e. turn signal based single lane change. Bringing AP2 up to its promised level of capability is not merely a technical challenge but a regulatory matter, i.e. time.
That's a variable that makes the relative cost benefit of AP2 an uncertain one for me, at least for the next 3-5 years. AP1 had originally promised capabilities that have not materialized yet, so AP2 will likely have its own challenges. AP1 is already extremely good, and AP2 costs twice as much and will have to perform that much better, very quickly, as the technology develops fast.
Seen solely from a technological+regulatory perspective, at the rate at which ML and computer vision capabilities are evolving, AP3 will be out by the time at least Level 3 or partial Level 4 capability is stable. At the rate at which NVIDIA is working, they'll supplant their PX2 (the heart of AP2) engine quickly, and Tesla might come up with their own in house design for AP3. PX2 was released at CES 2016. NVIDIA has already supplanted released the Xavier platform at CES2017. Tesla speaks of moving to their own platform in future. This represents twice changing vendors between AP generations - from Mobileye for AP1 to NVIDIA for AP2, and in future potentially an inhouse solution.
Therefore, like with a decision over buying phones, I will skip AP2 and probably get another Tesla when AP3 is out and L3/partial L4 capability is stable. That's the best value proposition for me - I have the best L2 capability car out there now as AP1, and I'll refresh when AP3 is stable.