Hype vs. Reality in AI Agent Economy: 6.4 Million-Payment Study

Financial commentary and industry events have been united in one message: self-governing AI programs are poised to take over finance and invest large sums without human guidance.

Research from BeInCrypto Intelligence, produced jointly with AWS, Cloudflare, Chainalysis and Zoomex, points to a major disconnect between the autonomous finance vision and the real-world actions of these automated systems.

The State of AI Agent Payments 2026 monitors real payments and AI-driven monetary tools. Its conclusions directly contest the widespread excitement around the topic:

• 90.8% of transactions in a 35-day sample were worth under a cent. Across 6.4 million payments, only $119,947 moved. Most activity involved API calls, computing resources, and automated data requests.
• Institutional infrastructure lags far behind micropayment systems. On the report's 100-point scale, micropayments scored 60.7. Institutional financial processes managed just 12.0.
• Human oversight still dominates financial choices. Of 19 AI tools examined, 14 operated within preset spending or trading limits. AI-labeled interfaces accounted for less than 0.55% of Hyperliquid's trading, and Zoomex order-preparation workflows required human approval.

Half of the 12 institutional workflows assessed could start or finalize payments within set boundaries. Whether companies will entrust agents with significant capital is still an open question.

BeInCrypto asked three sector executives about the barriers between micropayments and meaningful capital deployment.

Bigger Payments Need Safer Custody

Marcel Robert Harmann, founder and CEO of THORWallet, expects more agent-driven trading in the coming year. Platforms such as Hyperliquid already offer APIs for automated execution. The core challenge lies in controlling the funds.

“Microtransactions, where agents pay for data or compute, will continue to dominate the transaction count. Volume will likely increase as more actual trading runs through agents.”

Harmann points to threshold signatures as a way to reduce the risk of giving an AI wallet unrestricted access. Instead of a single key, a payment could require authorization from separate participants, including a signer that verifies policy compliance.

“With threshold signatures, the agent can hold one key share and an independent policy signer can hold a second one. We can adapt our workflows to the new situation.”

That approach could let agents initiate larger transactions without granting them full control over funds.

Harmann also predicts that regulators will adopt automated monitoring as transaction volumes grow. Humans cannot review every machine payment individually, he argues. Oversight systems will need to flag suspicious activity at machine speed, with people retaining ultimate authority.

The Financial System Itself May Need Rebuilding

ZeruAI founder and CEO Ashutosh Sahoo believes the obstacles go deeper than how assets are held.

“I don’t expect a meaningful shift beyond micropayments until we begin building an AI-native financial system. Existing banking infrastructure, compliance frameworks and liability models were designed around humans and institutions, not autonomous economic actors.”

Today's agents typically work within fixed spending ceilings, trading mandates, and emergency stop mechanisms. Sahoo sees these arrangements as supervised automation rather than genuine financial independence.

“The real inflection point comes when we stop fitting agents into human financial infrastructure and start designing financial systems around their capabilities.”

His vision would reshape compliance. Rather than checking most payments after the fact, systems could enforce spending permissions and risk rules directly through their protocols.

“If code is law, much of financial regulation could become enforceable at the protocol level rather than through retrospective surveillance.”

That raises difficult questions about who writes those rules and who can change them. Sahoo also worries that machine-designed financial systems could eventually become too complex for humans to understand, even when their actions follow a consistent logic.

An Agent Can Follow Every Rule and Still Lose Money

HumanWrite Technologies founder and CEO Tamara Frater expects uneven progress. Buying computing power is straightforward; investment choices carry far higher stakes.

“The next 12 months may bring more experimentation, but I don’t think the market jumps straight from sub-dollar payments to broad autonomous financial decision-making.”

Frater is especially concerned about errors that happen within an agent's authorized permissions. A system could respect its spending limit while acting on misleading instructions or sending money to the wrong place.

“The biggest concern is an agent making the wrong decision while still operating within its authorized permissions. Spending limits reduce exposure, but they do not guarantee that the recipient, instruction or underlying information is correct.”

The report notes two incidents where a combined $281,200 reached attacker-controlled wallets before being recovered. Those cases demonstrate that capping a machine's spending does not settle the question of whom it should trust.

Frater expects financial institutions and regulators to employ AI for identifying suspicious activity as automated transactions multiply. But she insists that decisions must remain traceable to responsible individuals.

“The challenge will be making oversight faster without making responsibility harder to trace.”

The Next Test is Real Capital

The three executives differ on how AI payments might evolve. Harmann concentrates on wallet security. Sahoo calls for financial infrastructure built around autonomous agents. Frater stresses the risk of authorized but incorrect transactions.

What unites their views is the gap between processing a payment and being trusted to make a financial decision.

For investors, the volume of transactions is a poor measure of adoption. A stronger test is sustained commercial payment value and whether companies trust agents with less human involvement.

For now, the first wave of agentic payments proves that machines can handle pennies on-chain. The next phase must show whether firms will let them manage serious sums.