The World May Not Be Desperate for AI Agents, Yet

Mark Zuckerberg has a vision for the future of computing that sounds increasingly familiar.

Within the next few years, he expects billions of people to have personal AI agents that know their preferences, understand their goals, and can perform tasks on their behalf.

These will not simply be chatbots that answer questions. They will be agents capable of booking flights, making reservations, managing schedules, shopping, and eventually handling increasingly important parts of people’s lives.

It is an attractive vision, but it is also one that Silicon Valley has a strong financial interest in making reality.

The question is whether consumers are actually desperate for it. Take something as ordinary as paying a bill through M-Pesa.

You already know what you need to do. Open the app, select the relevant option, enter the Paybill or Till number, enter the amount, confirm the transaction, and wait for the SMS.

An AI agent could theoretically do all of this for you by simply instructing it, “Pay my KPLC bill,” and it could identify the account, confirm the amount, initiate the transaction, and ask for approval before sending the money.

Technically, that is impressive; however, how much of a problem has actually been solved?

For most people, paying an M-Pesa bill already takes a few minutes. The process may be slightly tedious, but it is also familiar, predictable, and relatively easy. Shaving those few minutes off does not fundamentally change how people manage their finances.

The same question applies to many of the examples used to sell agentic AI. An agent could order your shopping from Carrefour, book a Bolt, schedule a doctor’s appointment, or reserve a table at a restaurant.

These are mostly tasks people already know how to do. The fact that AI can do them automatically does not necessarily mean people urgently need it to.

The AI industry is also remarkably confident about what can happen within five years.

Nvidia CEO Jensen Huang has argued that if artificial general intelligence is defined as an AI capable of passing essentially every human test, it could arrive within five years.

OpenAI’s Sam Altman and Anthropic‘s Dario Amodei have similarly suggested that AI systems capable of rivaling top human professionals could emerge within this decade.

The predictions extend beyond AGI.

AI companies and investors expect autonomous software engineers to handle much of the development process, while AI systems speed up scientific and drug discovery.

They also expect one-person companies to run with teams of AI agents and humanoid robots to take on work in factories and warehouses at scale.

There are good reasons to expect AI to become much more capable. It is also worth remembering that many of the people making these predictions are the same people building and investing in the technology needed to make them happen.

Meta, Nvidia, OpenAI, Google, Anthropic, and others have billions of dollars riding on AI becoming a foundational layer of computing.

That does not make their predictions wrong. It does mean they should be treated as forecasts from participants in the market, rather than neutral predictions about the future.

There is already evidence that consumers are interested in AI assistance without necessarily wanting to hand over control.

Booking.com found that 89% of consumers wanted to use AI in future travel planning, but only 12% were comfortable allowing AI to make decisions independently.

Expedia has found that almost 70% of travelers preferred booking through a trusted travel brand instead of an AI chatbot or agent. Two-thirds said they would not trust an AI assistant to buy or book something for them.

Essentially, people are comfortable asking AI to find them three good hotels but are less comfortable giving it the authority to book whichever one it thinks is best.

The real value of an agent is unlikely to come from automating individual five-minute tasks. It could come from managing the complicated web of tasks that people struggle to coordinate.

Picture an AI that does your calendar, recurring bills, travel plans, work commitments, and household responsibilities.

Instead of waiting for you to tell it what to do, it could identify conflicts, suggest solutions, and handle the routine steps needed to resolve them.

The killer application for AI agents may therefore not be task automation. It may be complexity management.

When Facebook became Meta in 2021, Zuckerberg positioned the metaverse as the next major chapter of the internet. He imagined people working, learning, playing, and socializing in immersive digital environments.

The metaverse did not become the next computing platform at the scale Meta envisioned.

While Zuckerberg did not promise that everyone would be living in the metaverse in five years’ time, Meta’s stated ambition was for the metaverse to eventually reach a billion people within the next decade.

The more useful lesson is that Silicon Valley repeatedly identifies a technology as the next major computing interface.

First it was the web, then social media, then mobile and then the metaverse. Now it is AI.

That does not make the prediction wrong. It does mean we should distinguish between a company’s vision of the future and evidence that consumers are already demanding that future.

Companies such as Meta are interested in AI agents because they could become the middleman between consumers and the businesses they interact with.

Today, you might search Google, visit a hotel website, and make a booking. With agents, the interaction could become:

You → AI agent → Hotel

The same could happen with shopping, travel, banking, entertainment, and countless other services. The AI agent becomes the intermediary. That makes it potentially one of the most valuable pieces of digital real estate in the world.

It also explains why technology companies are willing to spend enormous sums building these systems. The agent is not simply a digital assistant; it has potential to become the next platform.

People are clearly interested in AI that saves time, reduces tedious work, and helps them make decisions. What remains unproven is whether they want an AI that has enough autonomy to make those decisions for them.

The technology industry can predict when AI will become capable of booking your flight. It is much harder to predict when consumers will decide that they no longer want to book the flight themselves.

The main question for AI agents, therefore, is not whether they will become capable enough to run parts of our lives. It is whether our lives contain enough friction that we will actually want them to.

Silicon Valley is spending billions trying to solve the first problem, but the second one is still a consumer question.