We are entering a new age in which apps will no longer be needed for many tasks. George Maier predicts that AI agents will become the primary way to interact with digital products and services. This means that we may need to rethink the architecture, economics and governance of digital markets and the businesses that operate within them.
Apple’s App Store launched alongside the iPhone 3G in the summer of 2008, and with it came the beginnings of a massive market transformation towards the dominance of the app. Just as the dot com bubble before it had led to a surge in venture capital flowing into the web-business market, “disruptive” on-device applications distributed through app stores absorbed much of the VC attention in the 2010s.
Many of the prominent technology offerings we see today are app first, or even exclusively app-based. At the same time, large language models (LLMs) are serving to accelerate coding and the subsequent production of apps. A quarter of startups in a recent cohort at the Y Combinator accelerator (YC) had almost entirely AI-generated code bases.
AI will displace apps
However, the prediction I want to develop here is not the dwindling of app dominance due to the saturation of the marketplace. Rather, we are likely to see a more fundamental shift in how users interact with digital products and services, away from using apps for many common tasks, to using AI agents as the primary way to interact with digital products and services.
Offerings based on large language models (LLM) like ChatGPT already offer a diverse range of functions that traditionally would require multiple apps – to edit photos and write job applications, and for different aspects of users’ professional and personal life.
Now there is a new moment of acceleration that is likely to happen when AI agents can interact directly in a diverse range of digital marketplaces through middle layer APIs – a term which in this context refers to the technological mediators in transactions between users’ AI agents and the fragmented world of service providers.
The case of parking apps
To explain how this new paradigm might look, we can explore the example of parking apps. In the UK, drivers who frequent a wide range of parking facilities will likely have a selection of parking apps downloaded to their phone. Different facilities may require a user to pay via RingGo, PayByPhone or one of the many other apps with a market share at present. Generally, a facility will be tied to one particular app. This creates a frustrating experience for users who end up having multiple apps and accounts for services they never really wanted in the first place, when a simple payment facilitation is all that should be required to pay for parking.
Local authorities in the UK are working on the National Parking Platform, a solution that serves as a middle layer between the owners of the parking facilities and the apps that facilitate payments. Rather than having a direct relationship with one app provider, local authorities can add their facilities to the Platform, which allows payment through any participating app. The user is then able to choose one app for a much wider range of parking options.
AI agents
This sounds like a sensible solution. However, current advances in LLMs and the AI agents built upon them open up a new possibility: removing the need to have any dedicated parking app at all. With the National Parking Platform as an established and trusted middle layer, it is not beyond the scope of reason to imagine that a future AI agent would be able to interact with a middle layer like this directly.
For instance, an iPhone user might arrive at a new parking facility they have never visited before and ask a future version of Siri “can you pay for an hour’s parking here”? Siri could then use the device’s current location to find out who owns the parking facility and to interact via the National Parking Platform’s API directly, retrieving the price of the parking and any conditions attached. The user can confirm the payment using their face ID, with Siri facilitating the payment and securely providing requested information (such as the registration of the vehicle). In this scenario, the user never downloaded a parking app at all. The user’s on-device AI agent interacted directly with a trusted middle layer API to facilitate the exchange.
The middle-layer model
Frameworks like Apple’s app intents already allow developers to expose actions within their apps to Siri, so that a user can perform tasks without having to directly open the app. But what I’m suggesting here is a step further: the app itself would be no longer required, and Siri would discover and interact with exposed middle layer APIs on the internet.
There are some providers already offering a type of middle layer API service, such as Zapier and Make, although these remain fairly closed down, lack broad interoperability standards and are not specifically designed for agent-led interaction. What’s missing is a general-purpose open framework that allows AI agents to autonomously find, understand, and safely use these services on behalf of users.
It’s not difficult to imagine this type of middle layer model operating more broadly to book a restaurant, buy train tickets, order a taxi and get a hotel room. In instances where the user might want to “shop around” before making a choice (reading menus, looking at pictures of different hotel options), an interactive front end could be produced ephemerally by the AI agent in real time – in keeping with the design and experience preferences of the user. For example, a user might care about finding room in a hotel with good breakfast choices. Their AI agent could produce an interface in real time that allows them to compare hotel rooms based on detailed information about the breakfasts on offer.
From app-centric to agent-centric
While many companies are currently attempting to integrate AI features into their individual apps in an attempt to improve the user experience and add more functionality, we might consider a future where this ownership of the user-facing front end is lost for many businesses all together. The focus of product and service providers would turn to building a comprehensive middle layer API that can interact with a user’s AI agent of choice.
The future of digital markets
This potential shift, from an app-centric to an agent-centric model, raises a host of complex, interlocking questions. If AI agents begin to mediate many of our digital interactions, we will need to rethink the architecture, economics and governance of digital markets and the businesses that operate within them. A few of the most pressing considerations include:
Trust and identity
If AI agents are acting on our behalf, how can systems verify that they truly represent us? Similarly, how will users ensure that the APIs and digital services their agents interact with are legitimate? New standards of authentication and accountability will be essential, both for people and platforms.
Infrastructure and interoperability
For agents to function across a broad spectrum of services, we will need searchable, discoverable, and standardised APIs, a kind of “public index” for the machine-readable web. These APIs will also need to be flexible enough to handle varied agent requests and evolve with use.
Market power and participation
Agent-based ecosystems could either democratise access to digital services or entrench existing platform monopolies. Will smaller providers be able to compete? Will open-source and decentralised agents gain ground? Ensuring open standards and fair access will be critical to maintaining a healthy, pluralistic digital economy.
Business model realignment
If the user interface is increasingly owned by AI agents, many companies may lose their direct connection to customers. This could fundamentally reshape value chains, from pricing and branding to user engagement and data collection. Entire sectors may have to rethink how they generate revenue in a world where the app itself is no longer the gateway.
App persistence and exceptions
There are apps where the AI model of interaction outlined here doesn’t make as much sense, including where the purpose of the app is the end user experience or where regulatory and security pressures mean having a dedicated app is the easiest way to ensure compliance.
Embodied AI and augmented reality
Finally, as AI agents become embedded in spatial computing platforms, from smart glasses to in-car systems, our experience of digital marketplaces could become increasingly fluid, ambient and context-sensitive. We might no longer “open apps,” but instead live inside a web of AI-mediated interactions.
Sign up for our weekly newsletter here.
- This blog post represents the views of its author(s), not the position of Nutmeg, JP Morgan, LSE Business Review or the London School of Economics and Political Science.
- Featured image provided by Shutterstock
- When you leave a comment, you’re agreeing to our Comment Policy.
