AI Agents and the Future of Custom Apps
How AI is transforming app creation and user interfaces
1iPhone's legacy redefined
Steve Jobs' 2007 iPhone introduction framed three devices in one, foreshadowing today's AI convergence. The internet communicator aspect, least understood then, now manifests as AI agents. These agents combine multiple functions into single interfaces, much like the iPhone merged separate gadgets. The parallel extends to how both innovations initially confused audiences about their transformative potential.
2Messaging as AI's killer app
Just as messaging became smartphones' dominant use case, conversational interfaces now drive AI adoption. Meta's WhatsApp acquisition and failed Facebook Home experiment revealed the tension between unified interfaces and specialized apps. AI agents resolve this by maintaining conversational simplicity while accessing specialized functions when needed, without requiring users to switch between discrete applications.
3From 689 apps to infinite
Where smartphones required separate apps for each function, AI agents generate custom interfaces on demand. The author's Muse-created recipe app exemplifies this shift - built in 5 minutes during a dog walk. This represents the 'infinite app' paradigm where agents create disposable, personalized interfaces rather than requiring pre-built solutions for every possible need.
4Agents need computers too
Effective AI agents require computational resources akin to human users. Meta provisions each U.S. Muse user with a virtual machine (2-core CPU, 8GB RAM, 8GB storage), recognizing that AI needs its own 'computer' to function. This infrastructure enables agents to perform complex tasks like analyzing Instagram content or creating custom apps without taxing the user's device.
5The enterprise agent shift
Microsoft's Copilot evolution into a 'super app' with Autopilot features demonstrates enterprise adoption. Autopilot operates as a 'digital teammate' with its own cloud instance, identity, and permissions structure. This enterprise-grade approach integrates agents into existing workflows via Teams and Outlook, treating AI as a colleague rather than just a tool.
6From discovery to inspiration
The web's abundance shifted focus to discovery (solved by aggregators like Google). AI's abundance shifts focus to inspiration - determining what to do rather than how to do it. Agents become ultimate gatekeepers, controlling not just access to services but the very initiation of tasks. This inverts traditional models where apps competed for user attention.
7Custom UI generation arrives
Generative UI, predicted as the bridge to wearable computing, now emerges through agents. These interfaces adapt to immediate needs rather than offering fixed functionality. The author's recipe app demonstrates this - created specifically for their saved Instagram content, with no need for broader distribution or maintenance.
8Meta's hardware pivot
Meta Connect revealed glasses as Muse delivery mechanisms rather than standalone AR/VR devices. This positions Meta's hardware as always-available agent access points. The strategy mirrors how smartphones became internet gateways - the hardware matters less than its role in enabling the transformative software layer.
9The new platform war
Microsoft, Meta and Apple now compete to provide the operating layer for AI agents. Microsoft frames Copilot as 'a new OS for work', while Meta embeds agents in social ecosystems. Apple's approach remains unclear but will likely leverage its hardware integration strengths. The winner will control the abstraction layer between users and task execution.
10Permission architecture challenges
Enterprise agents like Autopilot require sophisticated permission systems, auditing and governance. These mirror human employee management structures, raising questions about AI accountability. The technical challenge shifts from functionality to compliance - ensuring agents operate within organizational policies while maintaining productivity benefits.
11Rate limiting realities
Current agent capabilities face platform-imposed constraints, like Instagram's rate limits on content access. These friction points reveal where legacy platforms may resist agent integration. The tension parallels early mobile web struggles, where desktop-oriented services initially failed to adapt to smartphone paradigms.
12The interface singularity
We're approaching an inflection point where the concept of discrete 'apps' disappears. Just as the iPhone merged devices, AI merges interfaces into contextual, just-in-time experiences. This doesn't eliminate specialized functionality but abstracts it behind conversational interfaces that assemble needed components dynamically, fundamentally changing how we interact with technology.
Original article: Stratechery by Ben Thompson — AI Agents and the Future of Custom Apps