What Happens When Users Stop Looking for Apps?
OpenAI gave us a glimpse of the future at today's DevDay in San Francisco.
For the last 15 years, software discovery has followed a fairly familiar pattern. You realize you need something, you search Google , maybe check Reddit, Inc. , browse the Apple App Store, compare a few products, install one, create an account, learn yet another interface, and eventually get to the thing you actually wanted to accomplish.
Take something as mundane as managing a tennis league. Today, the process starts with “I need software for scheduling my tennis team.” What OpenAI showed at DevDay points toward a different interaction: “Help me manage my tennis team.” That sounds like a minor UX change, but strategically it is much bigger than that.
The reason is that the user is no longer searching for software. The user is expressing intent, and the system is deciding which capabilities are needed to satisfy it. If ChatGPT can discover a specialized tennis-league plugin in the middle of that conversation and connect it immediately, then the fundamental discovery object changes.
Google indexes information. The App Store indexes applications. An agentic platform can increasingly index capabilities.
That distinction matters because your product no longer has to be something users already know exists, remember the name of, search for, install, and learn. It simply has to be the best available capability for a particular intent, and the agent can handle much of the discovery and orchestration around it.
This also starts to change what an “app” actually is.
Most software today is designed around getting the user into its environment. You open the app, navigate the interface, find the relevant workflow, and then perform the action. That structure made sense when the application itself was the primary interface between the user and the capability.
Agentic platforms invert that relationship. The user stays in the environment they are already using, and that environment calls your software when it is useful. The app becomes less of a destination and more of a service that can be invoked at the right moment.
Imagine you build an excellent expense-management product. Today, you probably need to build a dashboard, navigation, onboarding, authentication, notifications, settings, search, perhaps a chat interface, and increasingly some kind of AI layer on top. But if ChatGPT or another agent already supplies the conversational interface, a surprising amount of that becomes redundant.
The core of your product might instead be a set of capabilities: classify an expense, find a reimbursement, generate a report, submit an expense. The surrounding UX does not disappear entirely, but the important part of the product becomes increasingly headless.
We have seen versions of this transition before. Websites became services, services exposed APIs, and APIs allowed software to be assembled from other software. The difference now is that the API consumer is not necessarily another developer building a predefined integration. It is an intelligent agent deciding dynamically which capability to use based on what the user is trying to accomplish.
That leads to what I think is the deeper structural change: applications may increasingly split into two layers.
Historically, most successful SaaS products have owned both the data and the interface. Salesforce stores your customer information and gives you the interface for working with it. GitHub stores repositories and gives you the interface for managing development. Google Drive stores files and gives you the interface for finding and manipulating them.
Agents can separate those roles. You end up with systems of record and systems of action.
Systems of record own durable state: the customer database, the files, the payments, the source code, the permissions, the history, the business rules. Systems of action sit above them, understand what the user wants, and coordinate whatever underlying tools are required to make it happen.
ChatGPT clearly wants to own more of that second layer, and that is much more strategically important than the fact that AI can draft emails or summarize documents.
Your CRM might still be Salesforce . Your payments might still run through Stripe . Your files might still live in Google Drive. But if the user increasingly interacts with all three through an agent, the underlying systems remain critical while their interfaces become much less important.
Instead of opening Salesforce, clicking through dashboards, filtering opportunities, and manually assigning follow-ups, you might simply say: “Show me the deals most likely to slip this quarter, explain why, draft follow-ups for the account owners, and remind me Friday if nothing has changed.”
The agent figures out which systems it needs, retrieves the relevant state, invokes the appropriate capabilities, and presents the result. Salesforce still has enormous value because it owns the data and business logic, but the Salesforce interface may no longer be where the user spends most of their time.
That has significant implications for where software companies should invest.
If agents increasingly own discovery and much of the interface, then things like navigation design, onboarding flows, and even portions of the application shell become less defensible. They still matter, but they are no longer where the deepest value necessarily lives.
The valuable layer moves toward proprietary data, unique workflows, domain expertise, reliable execution, permissions, integrations, and transaction infrastructure. In other words, the part of the product that actually does something difficult and hard to replicate.
There is also an interesting distribution implication here, especially for smaller software companies.
For years, distribution has favored incumbents because they own the brand, the installed base, the SEO, the enterprise sales team, and the customer relationship. A small developer building something narrowly useful had to somehow convince users to discover the category before they could even compete.
Intent-based discovery changes that equation. A tiny company does not necessarily need millions of people to search for “tennis captain software.” It needs to be the strongest available capability when someone says, “Help me manage my tennis team.”
If agents become a meaningful distribution layer, that is a very different market structure.
The question for software companies may gradually shift from “How do we get users into our app?” to “What capability are we uniquely good at providing when an agent needs it?”
The app is not disappearing. But it may stop being the destination.