TechFlowPost argues that crypto’s opening in the agent economy may sit outside payments, pointing instead to on-chain collateral and on-demand risk markets.
The article, written by Lao Bai, says the author spent half an hour reading a long essay by @Leozayaat and came away with a strong impression: the more interesting question for the agent economy may not be how agents pay, but where crypto becomes necessary once agents start transacting at scale.
Why the payments thesis is being questioned
For some time, a common view in crypto held that payments would be the clearest entry point for blockchain in an agent-driven economy. The reasoning was straightforward. Agents can hold wallets easily, and small-value transactions appear well suited to blockchain rails. That led to a broader claim that blockchains may be built not only for humans, but also for agents.
Lao Bai says that view is now being challenged. The piece points to recent comments from DragonFly’s @hosseeb, who had also discussed similar ideas at 2049. He now believes his earlier judgment was wrong, writing that agents do not need crypto, while humans do. In this context, the article says, “do not need” refers mainly to payments.
According to the piece, Visa and Stripe have both been actively building around agent payments over the past year, and the vast majority of those transactions have been completed through traditional card networks and the broader financial system. By contrast, crypto protocols such as X402 account for only a negligible share. The conclusion drawn in the article is narrow but clear: agents do need payments, but that does not mean they need an entirely new payment network.
A three-stage framework for the agent economy
Lao Bai summarizes the original essay as a three-stage progression for the agent economy.
Stage one: Agents buying from humans
At the first stage, agents act on behalf of humans to buy services such as APIs, data, and compute. The central problem here is how to let an agent make payments autonomously.
Stage two: Agents buying from agents
Once agents begin buying services from other agents, the article says the core issue shifts from payment to performance. The question is no longer just how money moves, but how one side can be sure the other side will deliver.
Here, the original essay proposes using bonds or collateral to constrain agent behavior. The logic is that agent identities can be created without much friction, which makes reputation easy to game. Collateral that can actually be seized is presented as a more credible form of discipline.
Stage three: Disposable Markets
Lao Bai describes this as the most imaginative part of the framework.
The example given is an agent that temporarily needs 1,000 hours of GPU compute to complete a task, but worries that the supplier may fail to deliver on time. Instead of relying on trust alone, the agent could create a market on demand around that specific risk: whether the supplier will deliver on schedule. Other participants could then price that risk and commit capital to take exposure to it.
In this model, the market does not need to be listed in advance, and it does not need prearranged market makers. If there is real demand, the agent can try to find capital willing to underwrite the risk. The article’s broader point is that financial markets themselves could become infrastructure that is called only when needed.
Three takeaways from the article
Lao Bai closes with three observations after reading the original essay.
First, on stage-one payments, the author says X402 currently carries some sense of being “falsified,” which is part of why Haseeb publicly acknowledged that his earlier call was wrong. Even so, the article stops short of declaring the track dead, noting that it is still early.
Second, on reputation in stage two, Lao Bai links the issue to what ERC8004 is trying to address. But, like X402, this area also looks partly “falsified” based on current data, the article says. The short version of the argument is this: an on-chain identity does not equal credible reputation, and a reputation record does not equal real transaction history. For agents, the cost of farming reputation is very low.
Third, Disposable Markets point to a different economic form altogether. Agents could create one-off markets whenever they need to price a specific risk and then seek capital willing to bear it.
The article says that, at least logically, this makes sense in a silicon-based world because the way such systems search for information, discover options, make decisions, communicate, and price outcomes differs sharply from human society. Their efficiency, Lao Bai writes, could be higher than that of humans by “N orders of magnitude.”
He also makes clear that no one knows whether this is how the agent economy will actually develop. It remains an extrapolation. Still, in his view, it is a logically coherent one that opens up a wider set of possibilities than a payments-only framing.
Grok Bot and Meta Muse are cited as possible accelerants
Near the end, the article mentions the recent launches of Grok Bot and Meta Muse. Lao Bai says those releases make him think the evolution of the agent economy could speed up, and that new forms of financial markets may emerge that go well beyond payments.
He adds that the original essay is worth reading in full, noting that it includes a range of economics and finance experiments conducted by overseas labs using agents.



