The AI agents are coming

Stay informed with free updates

Simply sign up to the Artificial intelligence myFT Digest — delivered directly to your inbox.

A decade ago, after selling Siri to Apple, the creators of the voice-controlled AI assistant had another big idea.

With their new company, Viv, they set out to solve a persistent problem for smartphone users: How to accomplish everyday tasks without having to juggle multiple apps. Wouldn’t it be nice if you didn’t have to toggle between separate travel, hotel and map apps when planning a vacation, and could rely instead on a piece of software that integrated them all?

Viv never made the breakthrough it was looking for and was later folded into Samsung’s AI assistant, Bixby. But in tech, good ideas seldom die: They just wait for advances in the underlying technology to make them possible. As with much else, that advance has arrived in the form of large language models.

Creating what is in essence a new layer of digital plumbing between apps and websites like this hardly sounds like the most sexy use of AI. But it could end up causing important changes to how people use technology and shifting the balance of power in the tech industry.

AI agents that act on behalf of their users are the fad of the moment. Giving them the power to operate across different apps, web sites and digital services could have far-reaching effects.

AI start-up Anthropic, for instance, recently showed off an AI system operating a computer screen in much the way a person would. The demonstration, which it called Computer Use, showed the technology plucking information from different sources to fill in an online form — the kind of routine, low-level task that occupies the days of many back-office workers.

The idea of using a software program to exactly replicate something that a worker does on their computer screen has been around in a different form for years. Known as robotic process automation, or RPA, it involves programming virtual “robots” to perform tasks that span different apps. The natural language capabilities of generative AI have given this idea a new lease of life. Anthropic’s technology is designed to operate a computer just as a person would, though the software isn’t very adept yet at doing common things on a computer screen like scrolling.

For many office workers, services such as these that replace routine on-screen tasks could be the first real manifestation of generative AI. Matt Garman, head of Amazon Web Services, described his own company’s latest efforts to automatically co-ordinate work between groups of AI agents so they can complete more complex tasks as “RPA on steroids”.

The closest thing to the Viv idea of integrating apps in the consumer world, meanwhile, has come from Apple. Known as App Intents, it would require developers to adapt their apps to work with the Apple’s AI, enabling the software to work between apps without the user needing to open them. 

The implications of this seemingly prosaic idea could be far-reaching. If an AI assistant or agent can automatically access whatever data or functionality you need to complete a particular task, you would open fewer apps, visit fewer web sites and tap into fewer digital services. In effect, this would concentrate a person’s digital activity into fewer places.

One result is likely to be a rush by app developers to ensure they remain one of the primary places that continue to control user attention and act as hubs for completing tasks, rather than risk seeing their apps relegated to subsidiary status.

Ultimately, a handful of all-purpose AI assistants, acting like automated superapps, might come to dominate. If so, it could suck the value out of many independent apps, as users no longer open them or shower them with attention.

This presents app developers with a dilemma. Opening their services up to the big tech companies’ AI might mean losing their direct relationships with users. But trying to stand apart could result in their being shut out of the new digital ecosystems that are likely to coalesce around AI agents.

For their part, meanwhile, the biggest tech companies will be able to ensure smooth integration between their AI agents and their own apps, giving people more reason to gravitate to their technology.

This will present a new challenge to antitrust regulators. Just as they are getting to grips with the way that the largest tech platforms direct users to their in-house service — a practice known as preferencing — a whole new layer of technology could emerge that ties technology users even more tightly into Big Tech’s digital universes.

richard.waters@ft.com