I mean, they moved it from a deterministic set of actions to an AI-pilled LLM nightmare with probably a paucity of all test cases covered given the near-infinite permutations and actions available.
A smarter release would have been extremely limited in scope, heavily tested, and aimed simply for feature parity. Or even scoping it to just one section, like editing/adding/updating calendar events and all associated functionality. Or even exploring routing to much-less hallucinatory, more performant models like CNN for certain tasks instead of just throwing everything at probabilistic model and hoping for the best.
But that's not the way of AI-pilled middle manager thinking now.
A smarter release would have been extremely limited in scope, heavily tested, and aimed simply for feature parity. Or even scoping it to just one section, like editing/adding/updating calendar events and all associated functionality. Or even exploring routing to much-less hallucinatory, more performant models like CNN for certain tasks instead of just throwing everything at probabilistic model and hoping for the best.
But that's not the way of AI-pilled middle manager thinking now.