New research says AI agents aren't really spending money online yet

The pitch, repeated across the AI and crypto industries through 2026, is that autonomous "agents" will soon transact on their own — booking, buying, and paying without a human in the loop. New research suggests that isn't happening at any real scale yet.

That is the through-line of a report published today by Decrypt, headlined "AI Agents Spending Money Online? New Research Says Not Really." The framing is a direct pushback on the agentic-commerce narrative: despite the volume of announcements about agents that shop and pay, the underlying activity, per the research the outlet cites, remains marginal.

A caveat up front, in the interest of not overselling a thin story: the granular specifics of that research — who conducted it, the sample, the exact figures, and the methodology — were not laid out in what's accessible here. Cleartext will update this item if and when the underlying study is identified and its numbers can be checked against the source. We're not going to attach invented figures to a "research says" headline; doing that would repeat exactly the sin the story is about.

For market context on the day, Decrypt's own price ticker showed Bitcoin around $76,721, Ether around $2,475.52, and Solana around $99.74 as the piece went out. Those are quote-in-passing prices, not the subject of the story.

Key facts

• Decrypt reports that new research finds AI agents are "not really" spending money online at meaningful scale (Decrypt, 2026-09-13).
• The report's headline explicitly frames this as a counter to the agentic-commerce hype cycle (Decrypt).
• Prices shown on Decrypt's ticker at publication: BTC ~$76,721, ETH ~$2,475.52, SOL ~$99.74 (Decrypt price data).
• The research's provenance and specific figures were not detailed in the available reporting.

The real-world read

The interesting thing here is the direction of the skepticism. Most "AI agents will pay for things" coverage originates with parties selling the picks and shovels — agent frameworks, wallet and payment rails, and the tokens attached to them — for whom a large addressable market is the whole business case. A finding that says the spending isn't materializing yet is the rarer, less commercially convenient data point, which is a point in its favor.

It comes with two honest limits. First, this is secondary reporting; the actual study, its authors, and its numbers aren't spelled out, so "research says" should be treated as a lead to verify, not a settled fact. Second, "not yet at scale" is a snapshot, not a verdict on where the technology ends up. Take the headline as a useful corrective to the marketing, not as its own kind of certainty.

Opinion, and whose

The skeptical read — that agent spending is more announced than real — is Decrypt's framing, via the unnamed research it cites. Cleartext hasn't independently verified the study.

Sources

• Decrypt (Tier 2, reputable secondary), 2026-09-13 — reported that new research finds AI agents aren't meaningfully spending money online; also the source of the market prices quoted. The underlying research's authorship and figures were not detailed in the reporting.

This is news coverage, not financial advice.