Hoskinson Envisions Autonomous Agents Moving USDM Across Crypto
Cardano founder Charles Hoskinson has outlined a future where AI agents could independently interact with financial infrastructure, buy stablecoins and move capital between different blockchain networks. His latest example involves USDM, Cardano’s fiat-backed stablecoin, and an agent capable of using an ATM before deploying those funds across several crypto ecosystems.
Hoskinson said he could imagine a trading agent walking to an ATM, purchasing USDM and then moving the stablecoin across Ethereum, Solana, Hyperliquid and Cardano. The important part of the idea is not the ATM itself. It is the possibility that software agents could eventually control financial workflows without requiring a person to manually approve every individual transaction.
That would represent a different use of AI in crypto. Instead of an AI system simply providing trading signals or market analysis, an autonomous agent could potentially acquire funds, select a network, execute trades and use digital assets to purchase goods or services. The concept still involves major technical, security and regulatory questions, but the underlying infrastructure is beginning to take shape.
USDM is particularly relevant to this vision because it is a dollar-pegged stablecoin native to Cardano. Cardano describes USDM as fiat-backed and says each token is backed 1:1 by U.S. dollars held with regulated financial institutions. That makes the asset closer to digital cash than a volatile cryptocurrency such as ADA.
Midnight Could Become the Coordination Layer
Midnight is increasingly being positioned as part of this broader multi-chain environment. The network has already been working on cross-chain infrastructure, including interoperability with Cardano and EVM networks. Its August network update also highlighted the ability to support cross-chain applications while keeping certain information private through zero-knowledge technology.
Midnight City provides an unusual testing ground for this concept. The simulation includes autonomous AI agents operating inside a persistent virtual environment, giving developers a way to experiment with machine-driven interactions rather than simply building applications for human users.
The USDM ATM mentioned by Hoskinson appears to be part of that experiment. USDM’s presence inside Midnight City gives agents a recognizable financial object to interact with and creates a bridge between the simulated environment and the broader idea of autonomous economic activity. The Midnight City account said the ATM was made available for agents to test on the Midnight Preview Network.
Related: Cardano’s Charles Hoskinson Calls Justin Sun “One Direction—Up” in TRON Remarks
The bigger opportunity comes if agents eventually become capable of selecting between networks based on fees, liquidity, privacy and execution conditions. An agent could theoretically hold stable-value assets, move them to where liquidity is deepest and execute transactions without the user needing to understand which blockchain sits underneath.
That could also make interoperability more important than competition between individual chains. Ethereum, Solana, Cardano and Hyperliquid do not need to serve exactly the same purpose if an autonomous agent can move between them. Hoskinson has previously described Midnight as a potential coordination layer for agents operating across different networks.
There is still a large gap between that vision and everyday reality. An AI agent handling real money would need strict spending limits, secure key management, identity controls, reliable data and protection against malicious instructions. A system that can trade autonomously also creates new risks because a software mistake could potentially result in financial losses at machine speed.
The trading side is another major challenge. AI systems can analyze markets and execute transactions, but that does not mean they can consistently generate profits. Recent research reviewing automated investing systems found that evidence for durable, risk-adjusted performance remains limited, particularly once execution costs, changing market conditions and model failures are considered.
Related: Cardano Expands in Brazil as Exporters Turn to Blockchain for EU Compliance
For Cardano, however, the significance may extend beyond whether AI agents become profitable traders. If stablecoins such as USDM become assets that machines can acquire, transfer and spend, Cardano could become part of an emerging machine-to-machine financial economy. The value would come from actual transactions rather than simply having AI attached to a blockchain narrative.
Hoskinson’s Midnight City example therefore points toward a more interesting crypto use case: autonomous software becoming an economic participant. Instead of a person opening a wallet, choosing a chain and making a swap, an agent could eventually perform those steps according to a set of rules.
The immediate experiment is much smaller than that future. But putting a USDM ATM into an AI-agent simulation gives developers something concrete to test. If these experiments eventually connect autonomous agents with real assets, cross-chain liquidity and privacy-preserving infrastructure, the relationship between AI and crypto could become far more practical than today’s AI trading bots.
About The Author
