AI agents are emerging as the next major users of blockchain networks, driving high-frequency transactions and accelerating demand for infrastructure built for machine-to-machine payments and trading.
AI agents will lead on-chain activity, creating many more transactions than human users. Blockchain networks are changing with infrastructure built for autonomous, high-frequency payments and trading.
Machine-driven transactions are already on the rise, driven by AI agents and automated wallets. The next billion blockchain users won't be human.
That's the current discussion making the rounds now among developers and observers, after a series of events this month that have reframed the entire adoption discourse.
Web3 researcher Stacy Muur posted on X:
"AI agents will be the biggest on-chain users because humans cannot conduct thousands of transactions per minute.”
The message, published on July 10, came as the numbers were already beginning to back her up.
Six days ago, the Sui Network (SUI) hosted a livestream that was a mixture of carnival and stress test. Gamers connected with Google identities, played blackjack and drew on communal digital canvases.
Behind the humorous interface, AI agents traded, gamed and messaged at machine speed. The network registered a peak of 6,086,766 transactions per second, more than six times the stated target of one million TPS and roughly twenty times Sui’s previous controlled benchmark.
The mechanism, called Tunnels, opens an off-chain channel for high-frequency interactions before settling the final state back on-chain.
The base layer remained stable, even as the agents generated what network engineers described as a firehose of activity.
Over the past year, Sui has processed more than $1 trillion in stablecoin transfers, much of it driven by automated interactions. The July 4 experiment confirmed what many in the space had begun to suspect: the tests are no longer hypothetical.
Nexus research suggests AI agents could dramatically increase blockchain activity. It estimates that a decentralized exchange with 10,000 human users generates about 50,000 trades a day.
In comparison, just 100 AI agents performing market-making and arbitrage could generate about 10 million transactions over the same period.
Muur’s description of agents as “basically restless” has become a working summary among developers. AI agents can quote prices, rebalance portfolios, pay for data and computing, place and cancel orders, and settle transactions without waiting for human input.
That creates a steady stream of machine-to-machine activity that existing financial infrastructure, built around human-driven transactions, was not designed to handle.
Some payment rails are already being laid. Coinbase introduced x402, a protocol that allows software to make stablecoin payments on-chain in response to API requests.
Adoption of Coinbase’s x402 protocol was slow for months post-launch, with transaction volumes remaining negligible through mid-2025.
The turnaround comes after a pay-to-mint meme coin game on Base went viral, with transaction volume soaring 10,000% in a week. By the Q1 2026, cumulative x402 transactions had crossed 100 million, per blockchain analytics firm Chainalysis.
Chainalysis claimed the wallets behind much of that activity were not conventional retail users but instead demonstrated automated on-chain activity.
Most carried small amounts of USDC and executed narrow, repetitive tasks, a pattern consistent with software agents operating on predefined scripts.
The finding suggests that the protocol’s rapid growth was not fueled by an influx of human users but by bots systematically moving small payments on-chain.
The firm found that “tester-to-payer” conversion, or wallets going from dry runs to real payments, improved four-fold in six months, and weekly retention of agentic wallets continued to grow.
x402 is “beyond proof-of-concept. “Transaction volume is scaling,” Chainalysis said. The paper claimed that agents will also eventually need “payment infrastructure to support thousands of microtransactions per second, settled autonomously in real-time.” Those are capabilities that card networks and wire systems can’t provide.
Besides tests, finance is already flowing to real projects. In the first month of operation, Giza's autonomous yield allocation agent, ARMA, launched 25,000 bot instances.
It transferred $35 million worth of stablecoins and performed $5.4 million in on-chain transactions on Base. Those figures come from a single protocol, using designs that developers describe as early-stage.
Network engineers are now racing to keep pace. Sui’s Tunnels present one model: off-chain execution with cryptographic audit trails.
Nexus is developing a Layer-1 blockchain purpose-built for machine-speed throughput. Meanwhile, standards bodies are advancing identity infrastructure.
A proposal for ERC-8004, a registry for trustless agents, aims to pair with payment protocols like x402 to give bots verifiable identities and streamlined billing channels.
The challenge, multiple teams acknowledge, is that most existing blockchains were designed around human timing. One block, one signature, one confirmation. Agents don’t wait.
They demand cryptographic proof of execution and session-based architectures that can support continuous activity in the absence of congestion. Sudden sharding, specialized layer-2 networks and better data APIs are all critical priorities, all not roadmap items.
Muur warned repeatedly that the space is still in its “early innings.” Bots’ on-chain footprint is now in the billions of dollars, but the heavy, persistent activity of a fully evolved agent economy is still to come.
One developer, who asked to be identified by name, summed up the mindset of coders”
“The rails are the easy part. The trust and the scaling, those are the monsters under the bed.”
Across the sector, the conversation is now about the billions of machine-driven transactions that are arriving sooner than almost anyone expected.
No comments yet
Be the first to share your take when accounts launch.
