[Feature Request] Build an AI-native mobile OS instead of a screenless assistant

Feature request / product critique

The rumored direction of OpenAI hardware — a screenless, ambient, puck-like AI device — feels like a mismatch with what makes modern AI agents valuable.

AI's transformative capability isn't simply talking to users. It's acting on their behalf.

The useful agent is one that can navigate software, understand interfaces, interact with websites, manage authenticated sessions, fill forms, communicate, manipulate files, and coordinate workflows across services. A standalone screenless device risks recreating the architectural limitations of Alexa/Google Home: an intelligent interface trapped inside a sandbox and dependent on integrations or cloud APIs for meaningful actions.

Authentication is the key architectural problem

Consider:

"Find me the best flight to San Francisco next Thursday under $500 and book it."

A standalone device somehow needs Expedia/airline authentication, email/calendar access, payment credentials, identity information, and MFA.

Do users authenticate every service through companion software? Are authenticated sessions proxied through OpenAI infrastructure? Does every service need an OpenAI-specific integration?

Those approaches recreate infrastructure that already exists on the user's phone.

A smartphone already contains the user's authenticated digital environment: browser sessions, passkeys, biometrics, payment methods, applications, notifications, contacts, files, cameras, location, and network connectivity.

The agent should live inside that environment rather than outside it.

Recommendation: fork an existing Linux mobile stack

OpenAI does not need to invent a mobile operating system from scratch. Existing Linux/mobile projects already provide years of work around kernels, hardware abstraction, telephony, compositors, touch interfaces, and power management.

Fork an appropriate Linux mobile stack and make the browser/web runtime a first-class application platform.

Linux
    ↓
Mobile hardware / telephony layer
    ↓
Secure browser + Web/PWA runtime
    ↓
OpenAI agent runtime
    ↓
User applications and services

Instead of requiring a massive native application ecosystem on day one, make web applications installable as first-class applications:

expedia.com  → Expedia
spotify.com  → Spotify
github.com   → GitHub
reddit.com   → Reddit

Give web applications carefully designed native capabilities: notifications, offline storage, camera/microphone access, filesystem APIs, background execution, secure identity, and payment primitives. Native applications/compatibility layers can exist where necessary.

Give the agent secure OS-level primitives

An OpenAI-controlled OS could expose capabilities roughly like:

browser.navigate()
browser.inspectDOM()
browser.click()
browser.fill()

device.camera()
device.location()

files.search()
files.open()

notifications.read()

identity.requestAuthentication()
payments.requestAuthorization()

Sensitive actions remain protected by explicit permissions and biometric/user confirmation.

The agent doesn't need my Expedia password — I'm already authenticated locally.

It doesn't need unrestricted access to my payment credentials — the OS can request biometric authorization for a transaction.

It doesn't necessarily need Expedia to implement an OpenAI-specific plugin — it can operate the same web interface I can.

That is much closer to an AI-native personal computer.

A display isn't the enemy

AI should eliminate unnecessary interaction with screens, not eliminate screens themselves.

Most of the time I should be able to tell the computer what I want and let the agent execute it. But when confirmation, authentication, navigation, video, code, charts, documents, or other visual information is required, a display is extremely useful.

Voice should be an interface. It shouldn't have to be the entire computer.

Why this matters strategically

Apple controls iOS. Google controls Android. Deeply embedding an independent agent into either platform means operating inside someone else's permissions, APIs, commercial incentives, and legacy architecture.

OpenAI could instead control the complete software environment by starting from Linux + the web rather than reproducing 15 years of iOS/Android APIs.

The objective doesn't need to be "build another smartphone."

It could be:

Build the first personal computer whose primary user interface is an agent.

The phone form factor already contains almost everything such an agent needs: display, microphones, cameras, cellular connectivity, biometrics, NFC, GPS, battery, sensors, and — critically — the user's authenticated digital environment.

A separate screenless device that ultimately depends on a phone or cloud integrations for authenticated actions adds another abstraction layer.

Put the AI inside the computer instead. Fork mobile Linux. Make the web a first-class application runtime. Build secure agent primitives into the OS. Ship hardware whose complete software environment OpenAI can control. Then let the AI actually do things.

---

Filing this in 2026 so I can hopefully link back to it in Q1 2028. 😄