Some bots and scripts may possibly be adapted (it would have to be technically feasible, and that script/bot developers be sufficiently motivated). But then how could we (wikipedians) edit the message if it's not in an editable page on wikipedia but hidden in some bot or scipt code ? User message templates are changed quite frequently. And then again, users without the appropriate scripts would have to do everything manually. This would represent a significant disenfranchisement of those, which includes all unregistered users (and also many experienced users who don't find it useful to have plenty of scripts and prefer manually adding the templates). They would either have to do their best to write manually those complicated messages or good alternatives (which often require fetching diffs, article titles, dates, external links, etc), or they would add brief messages inferior in terms of editor guidance, or altogether stop notifying users, which according to wikipedia policy is necessary in various occasions. Even if some specific situations could eventually, at some point in the not too distant future, get covered by feature requests at bugzilla, that would never be reactive enough. We would seriously loose an enormous amount of functionality, even in user talk only.
And for the huge number of manually added inline templates, some of them I mentioned above, the only way to adapt would be to reverse to full manual, which would be a monumental increase in workload. Having to write down the wide variety of linking templates (citations, article, user, template links, helper templates at noticeboards, etc) would be excruciating, users would be discouraged. Let's not imagine what would become of AIV without AIV templates, RFPP without the RFPP templates, RFCs without inline links, TfDs without w:template:t1 and co - all of those templates require regular changes which excludes a bugzilla-mediated update process.
We would no longer be able to easily cite references on talk pages, which is crucial for verifiability discussions. So that we can take into perspective the extent of the loss, let me give as example one single citation template (one of the most used): cite web is transcluded almost 10,000 times in main talk namespace and a bit more than 2000 times in user talk namespace, excluding some sandbox pages. This isn't even the most used inline template on main talk pages, and on user talk pages, User has tens of thousands of transclusions.
And then as Arthur Rubin mentions, we could no longer give in-comment examples of templates when discussing them directly, which happens every day in all talk namespaces, just check out w:Help talk:Citation Style 1 or the thousands of transclusions in talk namespace of the template families infobox, navobox and co. And unlike for some talk page messages where the ability to subst: alone may be sufficient, try to subst one those templates, and then well :)