Is the App Model Going to be Replaced by AI?

The smartphone interface is approximately 20 years old. Carl Pei thinks that's a problem nobody is talking about loudly enough.

Speaking at SXSW in Austin this week, the co-founder and CEO of consumer electronics brand Nothing made a pointed argument: the way people currently use phones — lock screen, home screen, apps, app store, repeat — is not meaningfully different from the Palm Pilot era. The underlying logic hasn't changed. Only the hardware has.

His conclusion, delivered to founders and startups as much as to consumers: apps are going to disappear. Businesses whose core value lies within one should take that seriously.

The Argument Against the App Model

Pei's critique starts with something genuinely familiar. Getting coffee with someone requires navigating a messaging app, a maps app, a ride-share app, and a calendar app — four separate interfaces, each with its own logic, for a single intention. The phone doesn't know what you're trying to do. It just provides the tools and makes you operate them.

That's the gap Pei is describing. The technology has advanced considerably. The interaction model hasn't kept pace. Intentions still have to be manually decomposed into steps, and those steps still have to be executed across fragmented surfaces that don't communicate with each other.

His vision for what replaces it: a device that understands intention and acts on it, without requiring the user to navigate the execution. "I know you very well, and if I know your intention, I just do it for you" — that's the operating principle he's describing.

Three Stages of AI-First Device Development

Pei outlined a progression at SXSW that maps the distance between current AI device capabilities and where he thinks the category is heading.

The first stage — AI that executes commands on a user's behalf, like booking flights or hotels — is already being tested by several companies. Pei dismissed this as "super boring." It's still the user commanding the system, just with fewer taps. The interface logic is the same; the execution is automated.

The second stage is where the proposition becomes more substantive. Rather than responding to commands, the AI begins learning a user's long-term intentions and proactively surfacing relevant suggestions. The example Pei used was a user who wants to be healthier and receives nudges calibrated to that goal, without having to prompt the system each time. The device operates on a model of the user's objectives, not just their immediate requests.

The third stage — the one Pei finds most significant — is a device that acts without being asked. A system that knows you well enough to anticipate what you want before you've formulated the request. He compared the concept to ChatGPT's memory feature, but extended to the full operating environment of a device: "When the system knows us so well, it will come up with things that we don't even know we wanted."

The Interface Problem Nobody Has Solved

The most technically specific point Pei made at SXSW concerns interfaces — and it's worth unpacking because it has direct implications for how AI agents and apps will coexist in the near term.

Current AI agents that operate inside existing smartphone infrastructure do so by navigating human interfaces: tapping through menus, scrolling through options, simulating the gestures a human would make. Pei's argument is that this is a transitional and ultimately inadequate approach. An agent mimicking human navigation on a human interface is inheriting all the friction that interface was built to manage. It's not a new paradigm — it's the old paradigm with a layer of automation on top.

The future-proof version, in his framing, is an interface designed for the agent to use — not adapted from one designed for humans. That means APIs, structured data access, and interaction models built around machine-readable intent rather than visual navigation.

Nothing's own operating system already allows users to build mini apps through vibe coding — a signal that the company is experimenting with the space between traditional apps and agentic interfaces rather than simply waiting for one to replace the other.

What This Means for Founders and Marketers

Pei's SXSW comments were directed at founders and startups explicitly: if your app is where your core value lives, that value will be disrupted. The timeline is unspecified, and Pei acknowledged apps aren't disappearing imminently. But the direction is clear enough that the strategic question isn't whether to engage with it — it's when and how.

For marketers, the implications follow the same logic visible in agentic commerce, AI search, and conversational shopping. The surfaces through which consumers discover, evaluate, and act are shifting toward AI-mediated interfaces that don't require — and in some cases don't support — the traditional app or web experience. Attribution, behavioral data, and brand presence in those environments operate differently.

A device that executes intentions without requiring app navigation is also one that bypasses the engagement models most mobile marketing is built on. Push notifications, in-app experiences, app store optimization — each of these assumes an app-centric interaction model that Pei is describing as structurally obsolete.

Nothing closed a $200 million Series C last year partly on the strength of this vision. The fact that institutional capital is betting on it at that scale is its own signal of where the category is heading, independent of whether Nothing, specifically, is the company that gets there.

For marketing and growth teams thinking through what AI-first devices mean for their channel strategy, this is a useful moment to stress-test assumptions about where your audience will be — and how they'll be reached — three to five years from now. Winsome Marketing's team can help structure that conversation.