The Coming Era of Autonomous Economies: AI Agents as Workers and Consumers

What happens when AI can do more than answer our questions—when it can make decisions, spend money, and act on our behalf?

It could search for the right service, compare prices, choose an option, complete the task and pay the bill.

You would not need to sit in front of a screen and approve every step.

That may sound like a distant future. But how far away are we really?

AI is beginning to move in this direction. Instead of simply answering questions or generating content, AI agents are increasingly being designed to act on behalf of people and businesses. They can plan tasks, communicate with other systems, make decisions and, increasingly, participate in transactions.

This is a much bigger shift than simply having a smarter chatbot. We are moving towards a world where AI systems could become participants in economic activity—buying services, negotiating prices and interacting with other agents with limited human involvement.

The numbers already show that this shift has begun. Adobe Analytics reported that traffic from generative AI sources to U.S. retail websites increased 4,700% year-on-year in July 2025, showing how quickly AI is entering online shopping. Meanwhile, the x402 payment protocol, designed for programmatic machine-to-machine payments, had reportedly processed more than 165 million transactions by April 2026. Goldman Sachs Research estimates that around 300 million jobs globally are exposed to automation by AI—a measure of potential task exposure, not 300 million jobs disappearing overnight.

These are not simply predictions about what AI might do someday. They are early signs of AI becoming part of economic activity.

Economists Gillian Hadfield and Andrew Koh offer a useful way of thinking about this transformation. Rather than viewing AI only as smarter software, they examine AI agents as potential economic agents capable of planning and executing complex tasks with limited human oversight, and consider how they could interact with people, firms and one another.

If this continues, two questions become particularly important.

What happens to work?

The common fear is that AI will simply replace people. But I think the reality is more complicated.

Research from the Dallas Federal Reserve suggests that AI can substitute for tasks based on codified knowledge while complementing work that depends more heavily on experience and tacit knowledge. This creates an interesting divide: experienced workers may use AI to become more productive, while people entering the workforce could find some traditional entry-level roles harder to secure.

AI may therefore change how careers begin and develop, rather than simply eliminate every job.

What happens when AI starts spending money?

This is where the idea of an autonomous economy becomes even more interesting.

Companies are already developing infrastructure that allows AI agents to discover services and make programmatic payments. The x402 protocol, for example, is designed to let AI agents pay for APIs, data and other digital services without requiring a human to approve every individual transaction.

India is exploring this direction too. Reports in July 2026 said that NPCI is developing a proposed Unified Agent Protocol (UAP) that could provide a trust and authorisation layer for AI agents operating through UPI. The framework is still under development, so its final design and safeguards remain to be seen.

But giving AI the ability to transact creates a harder question:

Who is responsible when an AI agent makes a mistake?

If an agent buys something, signs a contract or makes a payment, how do we know why it made that decision? Who authorised it? And if many agents make similar mistakes at machine speed, could the consequences spread before humans have time to intervene?

The International Monetary Fund has highlighted precisely these concerns around agentic payments, including authorization traceability, opacity, correlated behaviour, cybersecurity and legal uncertainty.

Some countries are already experimenting with ways to address these problems. Singapore, for example, has tested AI-agent systems through a government-led sandbox and identified identity, authentication, permissions and user control as areas that need to evolve for more autonomous systems. Singapore has also seen live authenticated agentic payment experiments.

History offers a useful lesson. Businesses existed long before modern laws established clear rules around responsibility and accountability. AI agents may require a similar evolution. The goal should not be to stop them from participating in the economy, but to ensure that their actions are transparent, traceable and subject to meaningful human oversight.

The future may arrive faster than we expect. AI agents could eventually become as ordinary as websites and smartphones, quietly handling purchases, negotiations, payments and other economic decisions in the background.

The real question is no longer whether AI can participate in the economy. It already can.

The question I find more important is this: when machines can make decisions, spend money, negotiate and act on our behalf without asking us at every step, are we ready to decide not only what these agents are capable of doing, but also what they should be allowed to do?

That is the choice we are beginning to face—not someday in the future, but today.