AI and crypto: what is real today and what is still a pitch

AI and crypto meet in one practical place: software that acts on its own needs a way to pay and a way to prove who it is, and blockchains offer both. Some of that is working code today. Much of it is still a roadmap with a confident tone.

Our job here is to sort the two piles. Calmly, because both the hype and the doom on this topic are louder than the facts.

The fourth industrial revolution, as a lens

In January 2016, Klaus Schwab, founder of the World Economic Forum, published an essay and a book describing what he called the fourth industrial revolution. The first used steam to mechanise production. The second used electricity for mass production. The third used electronics and computers to automate it. The fourth, in the WEF’s framing, is a fusion of technologies blurring the lines between the physical, digital and biological worlds.

We use the phrase because it is a useful map, not because it predicts prices. AI is one of the technologies in that fusion. Blockchains are another. The question worth asking is not “is this a revolution?” but “which specific jobs does each technology actually do better?”

What an AI agent is

An AI agentSoftware that makes decisions and takes actions on its own. is software that does not just answer a question, it takes actions to reach a goal. Book the table. Compare prices. Pull data from several services and write a report.

The moment an agent does useful work on the open internet, it hits two very old problems. How does it pay for things? And how does anyone know whether to trust it? Card networks and logins were designed for humans with names and bank accounts, not for a program that wants to buy a single data lookup in the middle of the night.

Problem one: agents need to pay

Crypto, and especially the StablecoinA coin built to stay at one dollar. , is a natural fit for machine payments. A stablecoin is a token built to track a currency like the dollar, it moves between wallets at any hour, and a program can hold and send it without a human filling in a card form.

The most concrete example is x402, an open standard Coinbase published on GitHub. Its README calls it “an open standard for internet native payments.” The idea reuses an old, mostly unused web code, “402 Payment Required”. A server answers a request with that code and the payment details; the client pays and asks again; the server delivers. The standard says it aims to support many networks and forms of value, including stablecoins, tokens and fiat.

Why that matters: an agent could pay a tiny amount for one API call instead of signing up for a monthly plan. That is real code you can read today. How widely it gets used is a separate question, and the honest answer is that it is early.

Problem two: agents need to be trusted

If large numbers of agents from different companies start hiring each other, someone has to answer “is this agent who it says it is, and has it done good work before?”

One proposal is ERC-8004, titled “Trustless Agents”, a draft standard on Ethereum created in August 2025. Its abstract describes using blockchains so agents from different organisations can find, pick and interact with each other without prior trust. It suggests several ways to build that trust, matched to what is at stake: reputation from client feedback, re-checking an agent’s work, cryptographic proofs, and secure hardware.

Notice the word draft. ERC-8004 is a proposal under discussion, not a finished rulebook. It shows the direction serious builders are exploring, not a system you can rely on today.

Real today vs still speculative

Here is the sorting, as plainly as we can make it.

Real today

• Stablecoins move between wallets around the clock and programs can send them.
• Open payment standards like x402 have public code that developers can run.
• AI tools help people read contracts, summarise documentation and spot suspicious messages.

Early or experimental

• Agents paying each other for services at scale.
• On-chain identity and reputation for agents, via drafts like ERC-8004.

Speculative

• Whole “agent economies” running with little human oversight.
• Tokens whose main value proposition is having AI in the name.

When someone pitches you the speculative pile as if it were the first pile, that gap is where your money tends to go missing.

The risks that are already here

You do not need to wait for the future to meet the downside. AI makes scams cheaper to produce: smoother messages, convincing fake profiles, cloned voices. The rule is unchanged. The FTC says only scammers guarantee profits or big returns, and that crypto payments typically are not reversible.

“AI trading bots” deserve their own warning. A bot that promises steady returns is a pitch, not a product. And an agent you give wallet access to can spend whatever that wallet holds, so treat it like handing a stranger your card.

How to think about it without the noise

Ask three questions of any AI-plus-crypto pitch. What does the blockchain do here that a normal database could not? Can I see working code or a product, or only a roadmap? And who holds the keys if something goes wrong?

If the answers are clear, you may be looking at something interesting. If they are vague, the FTC’s warning about big claims without details applies just as much to AI as to anything else.

What to do next

• Understand the payment rail agents rely on in stablecoins explained.
• Get the big picture of ownership online in Web3 explained.
• Learn the patterns behind AI-polished scams in how to spot crypto scams.