The shift no one is naming clearly enough
Most of the conversation about AI in product is still about features dropped into existing apps — a chat panel here, a summarization button there. That framing misses the larger move. Apple Intelligence, Gemini, and the next wave of OS-level agents are not new features. They are a new substrate. They sit above every app on the device, interpret intent in natural language, and execute outcomes by chaining together capabilities the user never has to open.
In an agentic OS, the app is no longer the destination. The agent is. The app becomes a service the agent calls.
Three eras of consumer software
Each generation of consumer software has collapsed a layer of friction the previous one took for granted. The current shift is the largest of the three.
- The Installed Era. Software lived on the machine. You launched a program; the program owned the screen. Workflows were vertical and self-contained.
- The App Store Era. The phone collapsed software into discoverable, single-purpose apps. The home screen became the OS. Workflows fragmented across icons, but the user still navigated the surface manually.
- The Agentic Era. The home screen recedes. An OS-level agent reads natural-language intent, holds context across surfaces, and calls APIs across apps to produce outcomes. The user no longer routes themselves through tools — the tools are routed for them.
The third era is happening right now. Most products are still architected for the second.
What the agent actually does
An agentic OS does three things an app store does not:
- Interprets intent in language, not menus. “Get me back to the lodge” replaces three taps in a maps app and a fourth in a ride-hailing app.
- Chains APIs across vendors. A single instruction can move through ride-hailing, dining, scheduling, and music without the user ever surfacing those individual apps.
- Decides on the user's behalf. Predictive, proactive, context-aware. The system anticipates the next action and pre-stages it before being asked.
This is what changes the product brief. You are no longer designing for someone hunting through your interface. You are designing for an agent that needs your capability to be addressable, semantic, and reliable enough to call without permission.
What stays visual
This is not the death of UI. It is the death of transactional UI.
Forms, menus, settings, multi-step booking flows — the things users tolerate but never choose — are exactly the things an agent can execute invisibly. What remains on screen is the part of the experience worth keeping there: immersive maps, ambient information, AR overlays, replay video, real-time environmental data, the things you actually want to see.
The question for every product team becomes simple and unforgiving: if the OS handles every transaction, what is the screen still for? If you cannot name what a screen is doing that an agent cannot, the screen will go away.
The API is the product now
In an agentic stack, the unit of value is not the screen — it is the capability. That has architectural consequences most product orgs have not absorbed.
- Modular over monolithic. Capabilities have to be addressable as discrete actions, not buried inside multi-screen flows.
- Event-driven over request-response.Agents need to subscribe to changes in state, not poll a static endpoint on the user's behalf.
- Semantic over syntactic. An API an agent can reason about needs intent-level descriptions, not just parameter lists.
- On-device first. Personalization and privacy push compute toward the edge. Cloud-only architectures will lose both latency and trust against on-device intelligence.
Roadmaps still organized around screens are roadmaps for the previous era.
Stacking: the concierge effect
The interesting consumer experiences in the agentic era will not come from any one app. They will come from the chains— sequences of API calls the agent stitches together on the user's behalf.
Imagine ending a workout and, with no manual input, the system queues a ride home, drops a recovery playlist, and confirms a dinner reservation that fits the new arrival time. None of those apps is the experience. The chain is the experience. The agent is the conductor.
The companies that win this layer will not be the ones with the most beautiful apps. They will be the ones whose capabilities are most easily summonedby someone else's agent.
Three working principles for designers right now
- Design the capability, not the screen. Every feature should have an answer to: how does an agent call this without the user opening the app?
- Make the screen earn its visual time. If an interface is doing transactional work, plan for that work to migrate. Reinvest the freed surface area into immersion, expression, and craft.
- Treat the API as customer-facing. In an agentic world, your API is the part of the product the user actually experiences — they just experience it through an agent. Design it accordingly.
Forecast: the Apple–Perplexity scenario
A CEO transition at Apple changes the strategic calculus on this thesis faster than the 2027 horizon implies. Cook's tenure leaned on incrementalism: Apple Intelligence as a polite layer on top of an OS Siri couldn't carry. A new chief executive walks in with a different mandate — and the cleanest, fastest move to reset the agentic substrate is acquisition, not internal build.
Perplexity is the asset that fits the brief. They've shipped what Apple has not: a real-time, citation-grade answer engine that already chains web context, model reasoning, and follow-up actions in the way an OS-level agent will need to. Folding that capability into the iPhone, the Mac, the Vision platform, and the rumored ambient hardware (the home device, the glasses) would compress two product cycles of catch-up into a single announcement.
It would also reframe Siri from a feature inside iOS into a cross-surface agent powered by an answer engine Apple owns end-to-end — exactly the architecture this thesis describes. The market context cooperates: Google's antitrust pressure on the default-search deal makes vertical integration look inevitable; Apple's privacy stance makes a frontier-model partnership awkward; and Perplexity's growth sits at the right valuation window for a Big Tech check.
If the move happens, the inflection date for summoned, not opened pulls forward from 2027 to the next product event after the acquisition closes. The product teams that re-architect for agent-first capability calls before that announcement get a year of compounding advantage. The ones still building screen-first products read about it on the morning of.
The horizon
By 2027, the app stops being the destination and becomes a service in someone else's interface. The mobile-first decade was about making products easier to find. The agent-first decade is about making products easier to call. The teams that internalize that shift now — and re-architect from screen-out to capability-out — will be the ones the next generation of agents picks first.
We stop opening apps. We start summoning capabilities.
— EV