From where I’m standing right now in June 2026, it looks like we are moving towards a future in which AI is with you pretty much everywhere you take your smartphone or smartwatch or laptop, or at home in your bedroom with your smart TV. LLMs will power the “smarts” in those devices and for most use cases, you won’t be having to decide which model is best suited for what you want to do. Apple and Google are betting that with the data on your device (your name, age, gender, location, and everything else your phone knows about you) as context, their compact on-board AI will be good enough to be useful. Let’s assume they can make that happen in a couple years.
If this comes to pass, we may be looking at a looming collision between AI that works on device, has access to your personal data, and is smarter as a result, and a separate app that is just a gateway to its bigger, stronger frontier model that just hoovers up the data on your device and sends it to data centers for processing. Local versus API.PS - Technical note: Apple is actually doing this smartly. For basic things that can stay on device, its onboard foundation model will handle these tasks locally. But if the user asks Siri to do something more complex, it will gather what it needs and go do its work in a Private Cloud with a more powerful model (so ostensibly your personal data remains private). Personalized versus generalized. You expect different things from them because they have different strengths and weaknesses. One is more mundane, the other is more lofty, but mundane has actually proven to be difficult to get right. The mundane is just as difficult to nail than it is to, say, code a webpage, maybe even more so, because very few of us know how to write CSS but all of us know how to schedule an event in our calendar.
I’m much less disturbed by the mundane kind of “AI” than the lofty kind, the one that aspires towards AGI and consciousness and climbing Mount Everest, because it’s there. Though, maybe the mundane turns out to be more insidious. I’m sure the recent grads who were shouting down commencement speakers touting AI in their speeches were not doing so because they objected to models running on their phones making itineraries and booking tickets for summer vacations. It’s because doing the mundane work is how you’re supposed to move up the career ladder, and the C-Suite is pulling up the ladder because of AI promises to do that work for cheap. What’s not so clear is the cognitive atrophy that will result from all of this outsourcing.
The truth is, right now, AI hasn’t really penetrated that deeply into the lives of everyday people. The other day, my neighbor, who is in his 60s, told me he was getting a kick out of using ChatGPT to write lyrics for the songs he was noodling around with. Is this the quiet before the AI storm? There was a brief period in the 1990s when the minority of early adopters who had cellphones (me) were looked at funny by the majority who didn’t. Now, the people who are using OpenClaw running on a stack of Mac Minis to agentically orchestrate large swaths of their daily lives (not quite me) are being looked at in much the same way. It’s probably not going to feel like much of a storm. For most, it will be as quiet as an improved Siri that just works. Though, the ominous part will be the increasing low rumble of data centers driving up our electricity bills, lighting up the night sky.
