The iPad Pro Can Run a Language Model Now, It Still Won't Open a Terminal.
At WWDC 2026, Apple proved the M5 iPad Pro can run AI models on-device, then shipped iPadOS 27 with no additional desktop features. Can the iPad finally replace a laptop? Not yet, and that's a choice.
I’ve been keeping a stupid little dream alive for about six years, and I keep telling myself the next WWDC is the one that finally pays it off. The dream is simple. I want my iPad Pro to be my main computer. Not a companion, not a thin client to the Mac sitting in the other room, the actual thing I do my work on. I write on it at 6am before the day job, it’s the device I actually want to live on, and every June I sit down to watch the keynote half-convinced this is the year Apple hands the iPad the keys.
WWDC 2026 was not that year. Again.
I wrote two pieces about this, one when the iPadOS 26 beta dropped and one earlier this year about how a $599 MacBook can still do more than a maxed-out iPad Pro. Both came down to the same scorecard. Give me a terminal. Give me real browser engines instead of WebKit wearing four different costumes. Give me multi-user accounts, which is still the single most baffling omission on a device that families actually share. Give me a real Finder that understands ext4 and doesn’t choke on SMB. Give me clamshell mode, external display parity, the missing utility apps, and above all just give me a boring, unsexy Snow Leopard year where Apple fixes the shit that’s been broken for years instead of shipping new ways to break it.
So how’d iPadOS 27 do against that list? One item. Out of everything I asked for, exactly one thing showed up, and it’s the Siri overhaul, which I said in print I’d believe when I saw it. Fine. I see it. It’s real, it’s built on App Intents, and a Siri that can finally chain actions across apps by voice instead of demanding the exact magic phrase is genuinely useful. I’ll take the win.
Everything else is the same wall it’s always been.
No terminal. No Xcode. No real dev environment of any kind. No sideloading unless you happen to live in the EU. No JIT, so virtualization on the iPad is still the same useless twenty-percent-speed toy it’s been for years. Still WebKit only, and here’s the kicker, Apple actually moved further from the user-agent stuff this year, not closer to letting you run a real engine. No multi-user accounts. No clamshell. No external display that behaves like an actual second screen. No macOS app layer, no real Finder, no clipboard manager, no Activity Monitor, no Disk Utility. The community wishlists heading into this WWDC were asking for the exact same things I asked for back in August, which tells you everything you need to know about how much moved. Which is nothing.
Now here’s the part that actually gets under my skin.
The whole keynote, and most of the developer sessions after it, were about AI. And buried in there is a thing I can’t stop thinking about. iPadOS 27 lets my M5 iPad Pro run actual language models on-device now. Not phone-home-to-a-server features. Real local inference. I can load an open model and run it natively on the tablet, offline, no cloud, no token cost. Apple spent the entire year proving the iPad’s hardware is powerful enough to do serious computational work right there in my hands.
And then they still won’t let it open a Terminal.
The same device Apple is now bragging can run a multi-billion-parameter model locally is a device Apple has decided cannot be trusted with a command line. That’s not a hardware limit. It was never a hardware limit. That was the whole argument of both my articles, and I’ll be honest, I half-hoped I was wrong, because being wrong would’ve meant there was at least a technical reason for all of it. There isn’t. There never was. WWDC 2026 didn’t poke a hole in my thesis, it poured concrete around it.
This is a business decision dressed up as a product philosophy. Apple has the chip. Apple obviously has the engineering. What Apple doesn’t have is any interest in letting the iPad eat the Mac’s lunch, so the iPad gets to be powerful in exactly the ways that don’t threaten the laptop and powerless in all the ways that would. The M5 is allowed to think. It’s just not allowed to be a computer.
And maybe that’s the part I finally need to make peace with. There’s a rumor going around about a MacBook with a touchscreen, maybe even an Ultra, and the more I look at this year’s announcements the more that rumor reads like a confession. Because if Apple’s own flagship AI demos run the small model on the iPad and the big model on the Mac, if the whole toolchain still lives on the Mac and the iPad is just where the finished thing runs, then a touch MacBook isn’t Apple admitting the iPad can be a real computer. It’s Apple admitting the Mac was always going to be the one doing the heavy lifting, so they might as well make it pokeable and call it a day.
If that machine ships, I think I might actually stop. Stop trying, I mean. Trying has been the key word in this whole thing for years, and I’m starting to wonder how much of it has been me arguing with a company that settled this shit a long time ago and just never said so out loud.
Don’t get me wrong. I’m not throwing the iPad in a drawer. I still love writing on it at 6am with the house asleep and everything quiet. It’s the best piece of glass Apple’s ever made and probably the nicest device I own. But I think I’m done treating every June like it’s the year. The dream isn’t dead, exactly. I just finally believe Apple when they tell me, release after release, exactly what the iPad is allowed to be.
The hardware was never the problem. It’s the only part of this that’s ready.