> The next major device will definitely be AI-first.
Everything so far that has been named X First has been marketing woo woo, and in practice only meant "we're thinking about this use case a little more than before". Such as mobile-first, and cloud-first.
In either case, sure, it's very possible that device hardware will change. But in what way is hard to say. Will the on-device chips be more powerful to support local inference? Sure.
> Apple is currently trying to jerry-rig AI into their existing product, the iPhone [...] is bound to be a complete failure in the end.
Yes, kind of. The problem with all existing platforms including web is that they're build in a way that is adversarial to interop. Apps are siloed, and the only possible birds-eyed view is the OS itself. But, GUIs are not built for machine interop. Vision models to navigate UI will be flaky at best for the foreseeable future (and forget about voice, it's an extra modality at best and is way too limited). On web frontend, it's the same story. On backend, the web has been adversarial for a long time, with fingerprinting, rate limiting, anti-scraping, paywalling etc, which has been supercharged in the last year or two.
Essentially, the products and systems we use every day are a poor fit for interop with AI, so I suspect we'll see two parallel futures: (1) interop and semantic GUIs being integrated into platforms, web and app ecosystems (this is what MCP is IIUC). This will fail for the same reasons as web 2.0 failed (the adversarial nature of tech business models - opening up APIs is not incentivized), not to mention the investment required to build a new OS and (2) vision models to do tasks on behalf of humans with some mediocre agent-loop-thing on top of existing hot garbage pool of already flaky apps and sites. This won't necessarily fail, but it will mean platform- and large data owners (Google, MS etc) will yet again end up on top, since they control the access to the birds-eye view (much like Siri or Google Assistant). It is also the most noisy, flaky and data-intensive surface area to use for interop, meaning the products will be slow, bloated and feel like bonzibuddy for years.
Doesn't mean AI won't transform businesses and white-collar work. It certainly already does. But, the AI selling point for consumers (current ability - not "future potential"), is kind of like how Google Search and Maps was a decade+ ago. Sure, it provides amazing utility, but most of the time you're looking at memes, playing games and watching TV shows. AI in those products is mostly a continuation of ongoing enshittification.