I don't think FSD is going to be a thing this year or next unless you count token gesture sub-features,
I largely agree with you - the path to true FSD will be long, and we’re really only guessing if we try to say exactly how long, but it certainly won't be done and dusted by next year. More interesting is what the intermediate stops are along the way.
I’m not convinced there’s much scope for further improvement to autopilot type assistance. Already we have the nag problem: if you need to intervene every minute or two then you naturally remain alert, if it’s only needed at intervals of tens of minutes the temptation to lose concentration is too great and you need to be nagged. If it improves further to needing intervention only occasionally (maybe a regular commute on a well-mapped piece of motorway normally works without intervention), you not only have the problem of people tempted to break the rules, you also have the issue that even drivers who diligently monitor the AP have so little practice in intervening that they will get it wrong when called to do so.
So one approach is to stop at the current level of AP and wait a few years until FSD is fully baked. You might argue that’s the best way, but it’s pretty clear Tesla aren't going to do that.
Your suggestion that more effort should be put into monitoring systems is an interesting one - I think there's not only scope for better nag systems but also improvements in the whole intervention process. However, Tesla would probably say that this would be effort wasted in the long run and better to crack on with FSD.
So if the progress to FSD is incremental, where is the next step?
Tesla have been talking lately about on-ramp-to-offramp. It's not clear that they are actually talking about making that unmonitored: the only concrete thing they have mentioned is automatic lane changing. Automatic lane-changing just as an adjunct to existing AP is fairly useless and arguably a backwards step: telling the car to change lanes is one of the few things left to keep the driver engaged if the existing AP (lane following, TACC) is working well. However, it is something that's needed for the whole system, and one of the things that's perceived as a 'thinking' problem rather than a straightforward algorithm, so they may choose to showcase it in that way just to show progress.
In some ways it would seem easier to introduce the unmonitored capability from the other end. More complete versions of summon/parking should be some of the easiest to make safe. City driving at low speeds, though much more tricky in terms of the environment, has the advantage that it's lower risk: with AEB fairly effective at low speeds, the downside of getting FSD wrong is likely to be people honking at you and minor fender-benders, while getting it wrong at high speed probably leads to fatalities. Also at low speeds in traffic the car can simply stop where it is and say "sorry, I can't do this", waiting for the sleeping driver to wake up and take over, while at highway speeds it's got to cope with at least pulling over and probably driving for a considerable distance before it can do so if an insoluble situation pops up.