Nobody knows yet, and anyone who sounds certain is guessing. The bull case says AI agents become crypto's busiest users, because software cannot open a bank account but can hold a wallet. The bear case points to how small this still is: about $73 million settled by AI agents across about 176 million transactions in the year to April 2026, well under a dollar each.
A note on the legal and regulatory points here: they summarise public sources as of November 2026 and should not be relied on. Rules for AI agents are still being written, and some of what is described here is proposed or coming soon rather than settled. Check the current position, or ask a qualified professional, before acting on it.
How close are we to machines being most of the users?
By value and by working agents, a long way off; by raw transaction count on some payment rails, possibly closer than it looks. A crypto investment firm’s report counted those 176 million agent transactions between May 2025 and April 2026. It also counted more than 104,000 agents registered across at least 15 directories by the end of March 2026.
Registration is not the same as activity. A June 2026 academic study of ERC-8004, an Ethereum standard for agent identity, found that most registrations were placeholders. Depending on the network, only about 3% to 15% exposed a live service endpoint, meaning a working address another program could actually call.
So be careful with the word "users". On a blockchain, a wallet run by software looks like any other user, and one person can run thousands of them. A future where machines are "most of the users" could arrive on paper long before it means anything in practice. That is our inference from the data, not a measurement.
The bull case and the bear case, side by side
Both cases rest on real evidence; they differ on what the evidence means. Here they are on the same questions:
| Question | The bull case | The bear case |
|---|---|---|
| Scale | The head of a leading dollar-stablecoin issuer said in January 2026 that billions of AI agents will be conducting economic activity within about three to five years. | About $73 million in a year is tiny next to any mainstream payment network. |
| Why crypto | Agents cannot open bank accounts but can hold wallets and pay in cents, at any hour. | More than 98% of tracked agent settlements ran through one stablecoin: a single point of dependence. |
| Willingness | Crypto users are open: in one 2025 survey, 87% would let AI manage at least 10% of their portfolio. | A major research firm found three-quarters of US, UK and Canadian online adults uncomfortable letting an agent pay, even with spending limits. |
| Identity | Open standards now exist for agent identity and reputation. | The same study found reputation easily gamed; after removing fake feedback, 77.9% and 86.8% of rated agents on two of three networks had no valid ratings (15.8% on the third). |
| Markets | Agents can trade around the clock, which supporters argue adds liquidity. | The CFTC chair has warned that systems built on similar models may produce correlated trading. |
| Rules | Regulators are engaging in public, floating ideas such as "know your agent". | Liability is unresolved: who pays when an agent fails is still an open question. |
The research firm's analyst offered a useful yardstick. Phone-based mobile wallets took more than a decade to reach roughly a quarter of US online adults. Our judgement is that agent payments face a similar trust curve, though nobody can say how long it will be.
What would change for ordinary people?
If machines became the main users, people would mostly notice it in markets, payments and identity checks. Here is what each shift would look like, as we read the evidence:
- Markets. More of your counterparties would be software. Speed and data access would matter more, and the gap between a fast agent and a person tapping a phone would widen.
- Payments. Paying per request, a few cents at a time, would become normal for software. Most people would see it only as a line on a statement.
- Identity. Checks built for people would extend to software. In October 2026 remarks, listed in the CFTC’s index of public statements and remarks, the CFTC chair floated a possible "know your agent" standard and asked how regulators can tell whom an agent is trading for.
- Mistakes. Errors would travel faster. An agent can repeat a wrong action thousands of times before a person notices.
None of this requires agents to be smarter than people. It requires only that they are cheaper, faster and always on.
What does trust mean when the user is a machine?
Trust moves from the person to the permission: you stop asking whether you trust the agent and start asking what it is allowed to do. An August 2026 academic chapter on agent finance calls this bounded autonomy. On that view, the decisive design question is how tightly an agent's authority is drawn, not how clever it is.
That turns trust into things you can check. Who is the human or company behind the agent? What can it do without asking? Can its record be verified, or is it self-reported? Can it be switched off? Every agent acts for someone. The series has argued from Episode 1 that the someone still matters.
There is a second shift too. When agents deal with agents, trust checks have to be machine-readable. A person can read a project's lock or vesting schedule on a web page. An agent needs the same facts in a form it can query and verify before it acts. Much of the infrastructure for that does not exist yet, and building it may matter more than building smarter agents. That is our view, offered as judgement.
What could go wrong?
The worst outcome is not a single hack but a system-wide failure that nobody is clearly responsible for. Correlated agents could amplify a sell-off. Heavy dependence on one stablecoin concentrates risk. Fake activity could inflate the numbers people use to judge what is real.
Liability is the open wound. The CFTC chair's remarks asked directly who is liable if an agent goes rogue, and whether developers of trading agents would become subject to registration. Those questions have no settled answers yet.
Trust can also break suddenly. In a June 2026 UK survey, 60% of adults said they would stop using an AI shopping agent after a single mistake. One large, public failure could set adoption back years. That last point is our judgement, not a forecast.
The trust takeaway
Machines may become crypto's most numerous users. Whether that is good for people depends on whether every agent comes with limits, a verifiable record and an accountable human behind it.
Whatever agent you meet, ask:
- Who is the person or company behind it?
- What can it do without asking, and what are its limits?
- Is its track record verifiable, or self-reported?
- Is its activity real, or is it counting placeholders?
- If it fails, who pays, and can it be switched off?
Each of TrustSwap's products addresses part of this trust problem for people today. Team Finance provides token locks, vesting, staking and airdrop tools and has been used by 40,000+ projects. TrustSwap Launchpad runs vetted token raises, Panthea provides institutional custody, The Crypto App covers crypto news and portfolio tracking, and Alcyone is our strategy and advisory practice. Where this is going is versions of those checks that software can read, so an agent can verify a lock or a vetting record before it acts. That is a direction, not a shipped feature.