Paul Haddad on Mastodon:

I think Apple AI’s strategy should be do the thing only they can do, local AI but not the crappy little models made to run on phones. Focus on getting models running on only high end Macs with high (but reasonable) amounts of RAM first. Train it/them to work well for iOS/Mac coding and OpenClaw like work. Don’t worry about being “frontier” just good enough when running on >= 64GB of RAM.

This could work, but so few people have that much RAM. I don’t think Apple expected the cloud frontier to be so far ahead.