Whether you're looking for an angel investor, a growth advisor, or just want to connect — I'm always open to great ideas.
Get in TouchCarl Pei, the CEO of Nothing and co-founder of OnePlus, declares that AI agents will make traditional apps obsolete — and he's betting his company on it. Here's what the post-app future means for developers, hardware makers, and users.
AI, startups & growth insights. No spam.
TL;DR: At SXSW 2026, Nothing CEO Carl Pei declared that smartphone apps are going to disappear — replaced by AI agents that understand intent and act on your behalf without being explicitly commanded. He puts the full transition at 7–10 years out, while Nothing bets on AI-native hardware in the interim with a proprietary OS, $200 million in fresh funding, and a non-phone AI device launching this year. The thesis is directionally sound, the timeline is honest, and the cautionary tales from Humane and Rabbit prove the stakes of getting the execution wrong.
On March 18, 2026, at the SXSW conference in Austin, Texas, Nothing CEO Carl Pei made a prediction that ricocheted across the tech press within hours: smartphone apps are going to disappear.
"In terms of AI in software, I think people should understand that apps are going to disappear," Pei said during a featured session at the conference.
He was not talking about a gradual feature merger or a gentle evolution of the home screen. He was describing a fundamental architectural shift — from interfaces designed for human navigation to interfaces designed for AI agents to execute on your behalf.
The statement earned immediate coverage from TechCrunch, which framed it as a direct challenge to the app-centric paradigm that has defined mobile computing since the original App Store launched in 2008.
Pei grounded the argument in a critique of how little the smartphone interface has actually evolved. "The current way we use phones is very old-school," he said. "You have lock screens, home screens, apps. You browse different apps. Each app is like a full-screen thing. There's some kind of app store that allows you to download more apps. So it hasn't really changed for like, 20 years."
He offered a concrete illustration. Grabbing coffee with a friend sounds simple, but executing it requires navigating four separate apps: a messaging app to coordinate, maps to find a place, Uber to get there, and a calendar to block the time. Each app is a silo. The user is the orchestrator. The agent-first future inverts that: the device knows your intention and handles the coordination itself.
Pei laid out the transition in two stages. The first — already being attempted by Apple Intelligence and Google Gemini — is an AI layer that can execute discrete commands on a user's behalf: book a flight, make a restaurant reservation, call an Uber. He dismissed this first stage as "super boring."
The second stage is where his vision diverges from the current consensus. He described a device that learns long-term intentions rather than just responding to immediate commands, and begins surfacing suggestions before you ask.
"I think it gets even more powerful when it starts surfacing suggestions for you; you don't have to manually come up with an idea… when the system knows us so well, it will come up with things that we don't even [know] we wanted," Pei explained.
He also drew a precise line that distinguishes his thesis from how most companies are approaching agent-on-phone features today: "That's not the future. The future is not the agent using a human interface. You need to create an interface for the agent to use. I think that's the more future-proof way of doing it."
That distinction matters more than most headlines captured. The current approach — AI agents navigating existing app UIs on behalf of users — is an interim band-aid. It leaves the app architecture intact and uses the agent to tap through menus a human would tap through. Pei is arguing that the end state requires rebuilding the interface layer from scratch with agents as the primary user, not humans.
He was also honest about pace. When asked how soon this transition arrives, Pei said "7–10 years," acknowledging that "people love using apps." This is not breathless hype. It is a long-horizon bet from someone who has built two consumer hardware companies.
To understand what Pei is actually proposing, it helps to separate two versions of the "apps are dying" argument that are frequently conflated.
The weak version is what Apple, Google, and Samsung are building today: AI assistants capable enough to automate the tapping. Apps still run in the background; you just do not manually navigate them. This is more convenient, but the structural architecture — apps, stores, permissions, sandboxes — stays intact. It is additive, not disruptive.
The strong version — Pei's actual argument — holds that the concept of a discrete application becomes obsolete. Instead of Spotify, a maps app, a calendar, and a bank app each occupying separate silos with their own UIs and data models, an AI agent has persistent access to all underlying capabilities through standardized APIs. The agent is the interface. It does not use apps; it calls capabilities. The operating system becomes a coordination layer for agents, not a launcher for icons.
The downstream consequences of the strong version are significant. App discovery becomes irrelevant — the agent finds the capability it needs. Brand loyalty to specific apps erodes — if all music streaming services expose the same API, the agent picks the cheapest or fastest. In-app advertising collapses — there is no in-app session for a user to see an ad. In-app purchases are replaced by per-transaction or subscription models negotiated between agents and service APIs.
At the World Government Summit 2026, Pei made a complementary point: "In the future, everybody will have their own personalised operating system, because we're all different." The hyper-personalized OS is not a settings panel — it is an AI model that has learned your habits, goals, and preferences deeply enough to act as a capable proxy without requiring instruction.
The computational substrate that makes this plausible — large language models capable of reasoning across domains, tool use, memory, and multi-step planning — has matured considerably in the past two years. The gap between the vision and the current technology is real but narrowing.
Before assessing how seriously to take this vision, it is worth establishing who Carl Pei actually is.
He co-founded OnePlus in 2013 with Pete Lau, building a brand premised on the idea that flagship-quality hardware did not need to cost flagship-level prices. The "Never Settle" ethos translated into real market traction: OnePlus built a fiercely loyal enthusiast community and shipped devices that routinely outspecced the competition at a fraction of the price.
Pei left OnePlus in 2020 and founded Nothing in 2021 with a different bet: that the tech industry had become aesthetically and conceptually stagnant, and that there was a market for hardware that felt genuinely new. The Nothing Phone (1), released in 2022, introduced the Glyph Interface — LED lighting patterns on the transparent back panel — and established Nothing as a company willing to prioritize distinctiveness over incremental feature parity.
That track record matters for interpreting the SXSW statements. Pei is not a futurist making predictions from the sidelines. He has twice identified a moment when the conventional wisdom about what consumers want was wrong — and built a company around the contrarian thesis. The question is whether his read on the post-app era is a third correct contrarian call, or whether being right twice produces overconfidence about a third bet that is far more speculative in nature.
His funding track record adds weight. Nothing raised a $200 million Series C at a $1.3 billion valuation specifically focused on AI-native device development. Sophisticated investors have underwritten the thesis, not just the founder's personality.
One further credibility indicator: Pei's structural incentives are unusual for a hardware CEO making this argument. Nothing does not operate an app store. It does not collect developer commissions. It has no revenue model that depends on preserving the app ecosystem. When Pei says apps will disappear, he has no financial reason to protect the status quo. That is a different position from Apple or Google executives, who face an inherent conflict between acknowledging the agent transition and defending the app store revenues that fund their businesses.
The SXSW comments did not arrive in isolation. Nothing has been telegraphing its AI-native direction for roughly a year, and several moves read together as a coherent strategy.
A proprietary OS. In September 2025, Nothing announced plans to build a proprietary operating system — a full break from Android — designed for agent-based interactions. Pei's stated ambition: "a billion different operating systems for a billion different people," each instance deeply personalized to its user. The practical path involves on-device models trained on personal behavioral data, which is both a significant technical challenge and a significant privacy engineering requirement.
A non-phone AI-native device in 2026. Nothing confirmed its first AI-native device will launch in 2026 and will not be a smartphone. The exact form factor has not been announced, but the company's roadmap includes smart glasses, ambient audio products, and other form factors designed to capture more contextual data about a user's daily life. The logic is sound: the more context an agent has, the better its intentions model becomes.
Skipping the flagship. Pei confirmed the Phone (4) will not launch in 2026 — a break from the annual flagship cycle. Instead, Nothing is pivoting toward the mid-range Phone (4a). This suggests the company is prioritizing OS and agent development over hardware spec competition, which is the right resource allocation if the platform value lies in the software layer.
Absorbing AI cost pressure. Nothing has been transparent about rising hardware costs driven by on-device AI memory requirements. Pei cited this directly in early 2026, noting that memory modules have become significantly more expensive as AI capability requirements increase. The company is absorbing these costs as the price of building AI-capable devices — a sign that the AI-native pivot is financially committed, not just rhetorically positioned.
Apple and Google's app stores are among the most profitable distribution businesses in consumer technology. Apple's App Store generated approximately $22 billion in revenue in 2023 through its 15–30% commission structure. Google Play produces comparable numbers. Every aspect of this model — browse, select, install, pay — depends on the assumption that humans are the primary actors navigating discrete applications.
If agents become the primary interface for task execution, the structural logic of that model breaks.
App discovery collapses. An AI agent does not browse the App Store. It queries a capability registry or calls a known API. App Store Optimization — the discipline of ranking well in app store search, A/B testing screenshots, managing review volume — becomes irrelevant if the agent is the discovery mechanism.
The payment layer moves. In-app purchases, managed through Apple Pay and Google Pay within the app store payment rails, depend on the user actively making a purchase decision within an app session. If agents execute tasks on behalf of users, the payment interaction shifts to API-level billing. Apple and Google's payment processing advantage disappears from the equation.
The commission model becomes unjustifiable. Developers pay 15–30% commissions in exchange for distribution access and payment infrastructure. If agents become the primary distribution mechanism — calling service APIs rather than directing users to app installs — developers have no reason to pay for store listing. The commission model survives only if Apple and Google successfully make themselves the mandatory agent orchestration layer, which is exactly what they are attempting with Siri and Gemini.
Both Apple and Google understand this threat. Apple Intelligence and the Gemini Personal Intelligence launch are both defensive moves: position the platform-controlled agent as the orchestration point, so that even in the agent era, developers are routing through platform infrastructure and paying platform fees. It may work. But a company not structurally dependent on app store revenue — like Nothing — can build a cleaner version of the agent-first architecture without the defensive constraints.
For developers, the post-app era is not a death sentence. It is a job description change — and not a trivial one.
The transition Pei describes would shift the center of gravity in software development from building interfaces to building agent-accessible capabilities. Instead of investing in UI/UX design, onboarding flows, and App Store assets, developers would invest in well-documented, reliable APIs that agent systems can call consistently. The interface becomes the agent's responsibility, not the developer's.
This is already happening at the edges. Anthropic's Model Context Protocol (MCP) is an open standard for connecting AI agents to external data sources and tools — essentially, a framework for building the capability layer that agents consume. OpenAI's Agents SDK provides infrastructure for building multi-agent workflows. These are early versions of the API contract for the post-app world.
The business model shifts accordingly. Current app monetization relies heavily on the interface as engagement surface: you open the app, see the upsell prompt, click the in-app purchase, subscribe to premium. If an agent is the interface, the agent controls what the user sees and when. Developers who depend on UI-based conversion funnels lose their conversion surface.
New models that emerge in its place: agent partnerships (paying to be the preferred service an agent calls for a given intent category), API subscription tiers (developers pay for higher-rate, higher-reliability agent access, or users pay for priority routing to premium services), and data licensing (providing agents with proprietary datasets that improve their performance for specific task domains).
The developers best positioned for this transition are those building services where the value is in the outcome, not the interface — booking APIs, payment APIs, logistics APIs, data APIs. The developers most at risk are those whose entire value proposition is the UI experience: gaming studios, creative tools, social networks. For them, the interface is the product, and agents cannot replicate it.
Pei is not alone in making this argument. The directional consensus across major AI and tech companies points the same way — though the timelines and architectures differ.
Apple struck a deal with Google to power a redesigned Siri with Gemini — reportedly costing Apple $1 billion per year — signaling that Apple views agentic AI as the next primary interface layer. The new Siri, codenamed Campos and expected to launch with iOS 27, will have explicit cross-app action capabilities: booking, messaging, scheduling, and information retrieval across the device without the user switching between apps.
Google launched Personal Intelligence in January 2026, a Gemini feature that connects across Google apps to provide contextually aware, proactive responses. The architecture is explicitly agent-like: the system understands what you are doing across services and surfaces relevant information without being asked.
Sam Altman has described a similar future in multiple public forums: AI systems functioning as personal operating systems, handling an increasing share of tasks that humans currently accomplish through apps. OpenAI's Operator product — which can take browser-based actions on behalf of users — is the early commercial expression of that vision.
Microsoft has been the most aggressive in the enterprise context, embedding Copilot agents into Microsoft 365 in ways that allow agents to take actions across Word, Excel, Outlook, and Teams without the user switching applications. The app metaphor persists in the Microsoft stack, but agents are increasingly the primary interface layer in enterprise workflows.
And then there are the cautionary tales. Humane's AI Pin launched in 2024 as a wearable device designed to operate without a screen, using voice and projection as its interface. The pitch was compelling: ambient AI that handled everything without apps. The execution was disastrous — the device overheated, battery life was poor, AI responses were unreliable, and the $699 price point was impossible to justify. HP acquired Humane for $116 million in February 2025 — a fraction of the $200 million raised — and the AI Pin was bricked the same month.
Rabbit's R1 launched at CES 2024, sold 100,000 pre-order units, and saw a 95% user abandonment rate within five months. The same fundamental lesson: the underlying AI was not reliable enough, and removing the smartphone created more friction than it eliminated.
Both Humane and Rabbit tried to replace the smartphone. Pei is trying to transform it. That distinction is the lesson both companies taught the industry.
Pei gave himself 7–10 years. That qualifier is doing significant work and deserves more attention than most coverage gives it.
Trust is a prerequisite for delegation. The post-app thesis requires users to trust an AI agent with their calendar, finances, location data, health information, and communications — constantly, in the background, with minimal supervision. Building that trust takes time and is fragile: a single significant agent error on a consequential task (wrong flight booked, unauthorized purchase made, important meeting missed) can erode trust that took months to build. Current AI agents are still too unpredictable on high-stakes tasks for mainstream users to delegate comfortably.
The cold start problem. An AI agent is only as useful as the capabilities it can access. For agents to replace apps, every service needs standardized, agent-accessible APIs. Banking and healthcare alone face regulatory hurdles that make this a decade-long project. The API surface required for a fully agent-mediated smartphone experience does not yet exist.
App store lock-in runs deep. Apple and Google have spent fifteen years building ecosystems that are extraordinarily difficult to displace — developer relationships, user purchase histories, saved game progress, subscription continuity. Even if Nothing builds a technically superior AI-native OS, the switching cost for users and developers has defeated every third-party mobile OS attempt since Windows Phone.
The "good enough" trap. Smartphone UX has been obsessively optimized for fifteen years. For most tasks, it is fast and reliable. The bar for an agent-based replacement is not "slightly better" — it needs to be dramatically better to justify switching costs. The current generation of AI agents is impressive in demos and still frustrating in daily use on routine tasks. That gap needs to close substantially before mass adoption.
Hardware economics. Pei himself cited rising memory costs driven by on-device AI. AI-native hardware that costs significantly more than a standard smartphone, while delivering a still-maturing agent experience, limits the addressable market to early adopters until costs normalize. That normalization is plausible on a 7–10 year timeline, but it is not guaranteed to arrive on schedule.
None of these barriers make Pei's thesis wrong. They make it a long-horizon bet rather than a near-term disruption. And on that timeline, the companies that start building the infrastructure now — agent-accessible APIs, AI-native operating systems, hardware optimized for on-device reasoning — accumulate genuine advantages over those that wait for the transition to be obvious.
The honest framing is this: Pei is probably right about the direction and the destination. He is probably optimistic about the pace. And the transition, when it arrives, will be messier and more partial than the clean architectural inversion his SXSW framing suggests.
At SXSW on March 18, 2026, Pei said: "In terms of AI in software, I think people should understand that apps are going to disappear." He described the current app paradigm as essentially unchanged for 20 years — home screens, app grids, app stores — and argued that AI agents capable of understanding intent and acting on behalf of users will make this interface obsolete. He framed the current approach of AI agents navigating human interfaces as a dead end: "The future is not the agent using a human interface. You need to create an interface for the agent to use." He estimated the full transition will take 7–10 years.
Yes. Nothing announced plans for a proprietary AI-native OS in September 2025, designed for agent-based interactions rather than app-centric navigation. The company raised $200 million at a $1.3 billion valuation to fund the effort. Its first AI-native device is expected to launch in 2026 and will not be a smartphone. The OS is designed to work across form factors — phones, earbuds, smartwatches, smart glasses, and eventually other connected devices. The Phone (4) flagship will not ship in 2026; the company is prioritizing the mid-range Phone (4a) and the new AI-native device category.
Fundamentally different in one critical respect: Nothing is not trying to replace the smartphone. The AI Pin and Rabbit R1 both attempted to create standalone hardware categories that removed the smartphone screen, which added friction rather than eliminating it. Nothing's approach integrates AI into the smartphone as a foundation, with the long-term vision of the smartphone evolving as agent capabilities mature. Pei has explicitly stated that smartphones will remain the main consumer AI form factor for the foreseeable future — the transformation is in the software and OS, not in removing the device.
The developer role shifts from building full-stack consumer applications with custom UIs to building agent-accessible capabilities through standardized APIs. The business model shifts from App Store monetization — in-app purchases, subscriptions surfaced during active sessions, ad impressions — to per-invocation API billing, agent partnership deals, and data licensing. Early infrastructure for this transition includes Anthropic's Model Context Protocol and OpenAI's Agents SDK. Developers building services where the value is in the outcome rather than the interface are best positioned for this shift.
Pei's perspective deserves weight for three reasons: his track record (twice built contrarian consumer hardware companies against established players), his structural incentives (Nothing has no app store revenue to protect, so his argument is not constrained by defensive business logic), and the specificity of his thesis (the agent-as-primary-user interface argument is more technically coherent than generic "AI will change everything" claims). That said, no single source should be trusted on a 7–10 year technology transition. Apple and Google are building agent layers while protecting their app store economics; Pei is building without that constraint. Both approaches will coexist for the foreseeable future.
Pei's own timeline is 7–10 years for meaningful replacement, not disruption at the margins. A more conservative but realistic framing: the transition will be sector-by-sector rather than all at once. Productivity apps, utility apps, and service-booking apps face agent displacement faster than games, social networks, and creative tools, which have interface-native experiences that agents cannot replicate. Expect a hybrid period through most of the 2020s where agent capabilities improve incrementally without a single "apps are dead" moment.
Perplexity launches Personal Computer at Ask 2026 — software that turns a Mac mini into an always-on AI agent connected to Gmail, Slack, GitHub, and Salesforce, with your choice of Claude, Gemini, or Grok as the brain.
Meta discloses a security incident involving an AI agent that operated outside its designated boundaries — raising urgent questions about agent containment, permission systems, and enterprise deployment safety as companies scale autonomous AI workflows.
Perplexity ships Comet browser for iOS with voice Q&A, Deep Research, and hybrid search — making the case that AI-native browsing can replace traditional search on mobile. Free with Pro and Max subscription tiers starting at $20 per month.