The Rise of Agentic AI & The Machine Economy

The AI that the majority of us are familiar with and utilise are forms of generative AI — a "reactive content creator that produces a single output in response to a prompt." A complex, but one-step tool. On the other hand, an AI agent is a proactive artificial being that independently plans and executes a series of tasks for its human counterpart. An agent has its own set of capabilities that work together, allowing it to perceive a goal, formulate a plan, utilise tools to take action, observe what happened, and adjust — all in a continuous loop.

AI agents are already running in production at scale: agents that write code, run tests, process refunds, and resolve bookings, all without human intervention. From the vast capabilities of agentic AI, the world is witnessing the rise of a fully machine economy — an economy where machines are buyers, sellers, and intermediaries, all at once.

The Problem — Agents Cannot Transact Autonomously

"Buy me a pair of limited-edition sneakers" is a simple task that could be given to any agentic AI service. Agentic AI has no problem finding the sneaker, its size, price, and other details. The problem is that the agent cannot find a verified seller, nor can it verify if that seller is trustworthy.

Certain infrastructures already exist for agentic transfers, but they don't achieve full efficiency. Protocols such as x402 already allow agents to make payments and transact through crypto wallets. The real blocker is that agent A does not know that agent B exists, or whether to trust it. There is no universally adopted identity layer, discovery mechanism, or trust standard across agent ecosystems.

"Here's the shoe. It costs $__. Go buy it yourself."

The agent, every time

AI agents are projected to take over economy-wide transactions and handle trillions of dollars. Yet today, only 16% of US consumers trust AI to make payments autonomously. This isn't because agentic technology is absent — it's because the coordination layer that would make it efficient doesn't exist.

The Shortfall of Existing Solutions

Protocols like x402 and Google's emerging agent interoperability standards — A2A and AP2 — have made real progress. x402 enables agents to make instant stablecoin payments over HTTP, removing the need for human billing accounts. Google's A2A protocol provides a standard for agents to communicate, while AP2 introduces a framework for secure authorisation and payment execution.

However, these protocols assume that a counterparty is already known. Neither solves how Agent A discovers Agent B in the first place, nor does it solve the coordination process between unknown agents.

What Needs to Exist

For agents to transact fully autonomously, they need to have the judgement and verified trust of a human, but the speed of a machine. The model of what needs to exist is structurally similar to the path of the modern banking system.

Think about identity. In today's banking system each bank is given an 8–11 character SWIFT/BIC identifier. Banks can utilise SWIFT to access nearly every bank in the world through their respective codes. The way SWIFT revolutionised banking is exactly the change needed for agentic AI. We are not replacing wallets or payment rails — we are adding the missing layer above them.

While components of the five-layer infrastructure are emerging across various institutions, no existing solution currently unifies all five capabilities into a fully autonomous network. Natural developers or collaborators include large platform providers such as Google and Anthropic, and decentralised identity organisations like the Decentralised Identity Foundation (DIF).

References