Nothing CEO Carl Pei dropped a bold prediction at SXSW this week: smartphone apps are going to disappear, replaced by AI agents that understand your intent and act on your behalf. He’s wrong. Mostly. The timeline is fantasy, the scope is overstated, but the trend? That’s real.
The Timeline Is Hype, Not Reality
Pei’s talking like apps will vanish tomorrow. They won’t. Gartner predicts only 40% of enterprise apps will embed AI agents by the end of 2026—up from less than 5% in 2025. That’s significant growth, but it’s augmentation, not replacement. Microsoft reports 80% of Fortune 500 companies are using active AI agents, but again, for workflow automation and assistance—not as app killers.
Here’s what happened in 2025: 99% of enterprise developers experimented with AI agents, but mass adoption didn’t materialize. The developer consensus? Most agents are still fancy chatbots with extra API calls, not sophisticated enough to replace complex applications.
We’ve heard this before. Blockchain would replace databases. VR would replace screens. Voice assistants would replace keyboards. The hype cycle is exhausting. Reality moves slower.
Not ALL Apps—Just the Simple Ones
Pei says “apps are going to disappear.” He means some apps. Specifically, the simple utility apps that don’t provide unique value beyond data management.
Apps that might actually die: to-do lists, basic calendar scheduling, simple email managers, form-filling utilities, basic note-taking apps. If your app’s core value is “manage this data,” agents can do that without opening your app.
Apps that will survive: Creative tools like Figma and Adobe won’t disappear—try editing a video with voice commands only, then try it with a UI. See the problem? Games aren’t going anywhere. Apps with unique hardware integration—fitness trackers, specialized camera apps—provide value agents can’t replicate. Offline-first applications persist because agents need connectivity.
The reality most people miss: it’s not apps OR agents. It’s apps WITH agents. As one Hacker News commenter nailed it, “Productivity apps won’t disappear, just the need to open them will.” Apps become orchestration layers for agents, not casualties of them.
The Interface Problem
Pei argues we need to “create interfaces for agents, not humans.” He’s only half right.
Visual interfaces still matter for control, privacy, and transparency. Users need oversight—the ability to pause, redirect, or override agent actions. Creative work requires visual feedback. Try designing a logo, editing video, or modeling in 3D without seeing what you’re doing. Privacy concerns demand interfaces too: users need to see and edit what agents remember about them.
The future isn’t invisible interfaces. It’s better interfaces for orchestrating agents. Multi-agent dashboards are the 2026 reality—kick off tasks from one place, monitor agents across environments like your browser, editor, and inbox. It’s natural language input combined with visual control, not one or the other.
Where Pei IS Right
Pei’s not crazy. He’s just early. Very early.
Simple utility apps genuinely are vulnerable. If your business model is “act as middleware between users and APIs,” agents replace that middleware. If your app just aggregates information, agents aggregate better. The prediction that 80% of apps will disappear focuses on these utilities—and that prediction has merit.
Agent frameworks are exploding. Superpowers hit 4,000 GitHub stars in a day. Claude HUD gained 1,000 stars for just showing what AI agents are thinking. The developer community is betting on agents, building infrastructure, creating frameworks. The trend is real, even if Pei’s timeline is exaggerated.
Gartner’s 40% adoption rate by end of 2026 is substantial. The interaction model IS shifting from task-specific apps to intent-driven agents. This isn’t hype—it’s happening. Just slower than CEOs pitching AI-first visions would have you believe.
What Developers Should Actually Do
Don’t panic. Don’t rebuild everything.
If you’re a game developer, ignore most of this. You’re fine. If you built a to-do list app, start exploring agent APIs or pivot. Fast.
For everyone else, focus on integration, not replacement. Build APIs that agents can call. Ask where your app provides unique value beyond data management. If it’s visual control, emphasize UI quality. If it’s offline access, double down on local-first architecture. If it’s creative tools, invest in precision and feedback.
Watch the frameworks—Superpowers, LangChain, agent platforms—but don’t chase every new tool. The answer isn’t “become an agent developer overnight.” It’s “make your app agent-friendly while maintaining what makes it valuable.”
Timeline for action: 2026-2027, experiment with agent integrations. 2028-2029, consolidate simple utility features into agent-driven workflows. 2030 and beyond, reevaluate based on actual adoption, not CEO predictions at conferences.
Carl Pei sees the future of smartphones as launchers of outcomes, not apps. He’s partially right. The future is hybrid—apps that orchestrate agents, agents that enhance apps, and interfaces that balance automation with control. Apps aren’t dying. They’re evolving. And that evolution will take a lot longer than a single SXSW keynote suggests.
