The internet may be quietly splitting into two layers. One is built for people: the feeds we scroll, the news we read, the shopping carts we fill. The other is a newer layer in which AI agents hold accounts, move money and make decisions without a human in the loop for each step. Two developments from early 2026 show how far that second layer has already come.
Wallets for machines
On February 11, Coinbase launched Agentic Wallets, a product from its developer platform that gives AI agents their own crypto wallets on Base, the Ethereum layer-2 network Coinbase operates. The company describes it as wallet infrastructure built specifically for autonomous software, with authentication, spending controls and monitoring so that an agent can pay for services, receive payments and trade within limits set by its operator.
Within the first 24 hours, roughly 13,000 agents had registered wallets, according to figures circulated after the launch. Independent confirmation of that count is limited, but even an approximate number signals a shift: software is no longer just writing code and answering questions, it is acquiring a financial identity.
Bots that pay for themselves
Prediction markets offer a second example. A study posted to the arXiv preprint server found that a small group of highly automated traders extracted close to $40 million in arbitrage profit from Polymarket over 12 months by exploiting mispriced contracts, according to reporting on the paper. The strategies rely on speed and consistency rather than insight, which is exactly where software has the advantage.
Some commentators, including the author of Nate’s Newsletter, which first framed these developments as the start of a machine economy, argue that a few agents are now self-sustaining: they use trading profits to pay for the computing power that keeps them running. That claim is harder to verify than the aggregate numbers, but the mechanism is plausible, since both the income (on-chain trading) and the expense (API and cloud bills payable in stablecoins) can be handled programmatically.
What a two-layer web would mean
If agents can hold funds, earn and spend, they stop being merely tools and become economic participants. That has several practical consequences.
- Payments infrastructure changes. Protocols for agent-to-agent payments, such as Coinbase’s x402 standard for paying for web resources over HTTP, are being designed for machines that transact thousands of times a day in small amounts.
- Markets get faster and thinner. In prediction markets and decentralized exchanges, human traders increasingly compete against bots that close pricing gaps in milliseconds, leaving fewer easy profits for anyone else.
- Accountability gets murkier. When an agent with its own wallet causes a loss or breaks a rule, it is not always clear who is responsible: the developer, the operator or the platform that issued the wallet.
- Regulators have to decide what an agent is. Existing know-your-customer and anti-money-laundering rules assume a human or a registered company behind every account.
Separate worlds or one?
The open question is whether these two layers stay distinct, with humans and agents operating in largely separate economic spaces, or merge into a single system in which people routinely delegate spending and earning to software. The early evidence from Coinbase and Polymarket points toward merger: the agents are trading in the same markets and paying for the same cloud services people use. What is new is that the internet is no longer only a network for moving information. It is becoming an economy in which autonomous digital entities take part.
