For the last decade, our relationship with technology has been defined by the grid of icons on a glass screen. We have become expert navigators of the “app economy,” jumping from Uber to Expedia to Gmail in a fragmented dance of logins and loading screens. It is a seamless experience in some ways, but it is fundamentally a chore—a series of digital silos that require us to do the heavy lifting of coordination.
But a new shift is underway, moving us away from “app-centric” computing and toward “agent-centric” computing. The goal isn’t just to have a smarter assistant that can answer questions, but to have a digital agent that can actually execute tasks across different platforms without us ever opening an app. This is the promise driving the current wave of AI hardware, from the Rabbit R1 to the Humane AI Pin, and it represents the most significant challenge to the smartphone’s hegemony since the iPhone debuted in 2007.
As a former software engineer, I find the technical ambition here fascinating, if slightly premature. We are seeing a transition from Large Language Models (LLMs), which are essentially sophisticated predictors of the next word, to Large Action Models (LAMs). While an LLM can tell you how to book a flight, a LAM is designed to navigate the interface of a travel site and actually click the “purchase” button for you. It is the difference between a consultant and an employee.
The friction of the app store model
The current smartphone paradigm is built on “walled gardens.” Every company wants you inside their specific app because that is where they can control the data, the advertising, and the user experience. For the user, this creates immense friction. If you want to plan a trip, you might spend an hour switching between five different apps, manually copying and pasting dates and confirmation numbers.
The vision proposed by new AI hardware is the removal of this middleman. Instead of the user navigating the app, the AI agent navigates the app’s backend or its user interface on the user’s behalf. In this world, the “screen” becomes secondary. You don’t “open” Spotify; you simply tell your device to play a specific mood of music, and the agent handles the API calls and navigation in the background.
The hardware gamble: Rabbit R1 and Humane AI Pin
The industry is currently in a “prototyping” phase, characterized by bold, often polarizing hardware. The Rabbit R1, with its bright orange chassis and rotating camera, and the Humane AI Pin, a wearable that projects information onto the palm of your hand, are the first high-profile attempts to decouple the AI agent from the phone.
However, the transition has been rocky. Early adopters have reported significant hurdles that are familiar to anyone who has worked in hardware development:
- Battery and Heat: Running complex AI models locally or maintaining a constant cloud connection drains batteries rapidly and generates significant heat in small form factors.
- Latency: The “thought” gap—the seconds it takes for a prompt to go to the cloud, be processed, and return as an action—still feels sluggish compared to the instant response of a local app.
- Reliability: LAMs are still prone to “hallucinations.” In a chat bot, a wrong fact is a nuisance; in an action model, a wrong click could mean booking a non-refundable flight to the wrong city.
Comparing Computing Paradigms
| Feature | Smartphone (App-Centric) | AI Agent (Action-Centric) |
|---|---|---|
| Primary Interface | Touchscreen/Icons | Voice/Vision/Natural Language |
| User Workflow | Manual app-switching | Single intent, automated execution |
| Data Control | Siloed within individual apps | Aggregated by the agent |
| Hardware Goal | Maximized screen engagement | Minimized screen dependency |
The “iPhone Problem” and the ecosystem war
The biggest threat to these standalone AI devices isn’t necessarily their own technical flaws, but the incumbents. Apple and Google already own the hardware in our pockets. They don’t need to invent a new device to implement agent-centric computing; they just need to update their operating systems.
With the integration of more advanced AI into Siri and Google Assistant, the “agent” experience may simply become a feature of the smartphone rather than a replacement for it. If Apple can make a LAM work within iOS, the need for a separate Rabbit R1 or Humane Pin vanishes instantly. The startups are essentially betting that the smartphone is too bloated and restrictive to evolve, while the giants are betting that they can absorb the innovation and keep users within their existing ecosystems.
What remains unknown
While the technical roadmap is clear, the privacy implications are not. For an AI agent to work, it needs deep access to your accounts—your passwords, your credit card info, and your personal preferences. We are moving toward a model of “extreme trust,” where a single entity (the agent provider) has the keys to nearly every digital door in your life. Whether users are willing to trade that level of security for the convenience of not having to open an app remains the ultimate unanswered question.
The next critical checkpoint will be the rollout of the next generation of “AI-first” operating systems and the software updates promised by Rabbit and Humane to address early stability issues. As these agents move from novelty to utility, we will see if the glass screen remains our primary window to the world or becomes a relic of the “app era.”
Do you think you’re ready to ditch the app grid for an AI agent, or is the privacy trade-off too steep? Share your thoughts in the comments below.
Worth a look
