
For most of the past decade, the conversation around artificial intelligence and crypto stayed largely theoretical. Two industries, both moving fast, both attracting enormous capital, but mostly running on parallel tracks. That started to change in late 2024, and by early 2026 the overlap had become hard to ignore. MoonPay, the crypto payments firm that built its name on fiat-to-crypto on-ramps, is now positioning itself as the financial infrastructure layer for a future where AI agents don't just analyze markets but actively participate in them.
On February 24, the company officially launched MoonPay Agents, a non-custodial software layer built on top of MoonPay CLI, its developer-focused command-line interface. The product gives autonomous AI systems the ability to generate wallets, fund them through fiat on-ramps or crypto transfers, execute on-chain trades, and convert holdings back to fiat, all without requiring a human to approve each individual step. Less than three weeks later, on March 13, MoonPay followed up with a second announcement: a deep integration with Ledger, the hardware wallet maker, designed to let users sign off on AI-initiated transactions directly from a physical device.
MoonPay CEO Ivan Soto-Wright put it bluntly in the launch statement: "AI agents can reason, but they cannot act economically without capital infrastructure." The line is a bit pithy, but it captures the actual gap. Building a bot that can identify an arbitrage opportunity across three chains is a solved problem in 2026. Building one that can act on that opportunity, fund itself, execute the trade, and off-ramp the proceeds into a bank account without exposing private keys or requiring a human babysitter is not.
MoonPay Agents is designed to close that gap. The setup is relatively straightforward: a developer installs MoonPay CLI, a user completes a one-time KYC verification, funds a wallet, and grants the agent permission to transact within defined parameters. After that initial handshake, the agent can operate independently. Wallets are non-custodial and stored locally on the user's device using OS keychain encryption. Private keys never leave the machine. Spending limits and pre-execution transaction simulations serve as guardrails against runaway agents doing something unintended.
The product ships with 54 tools across 17 categories, covering most of what a developer building a financially active agent would actually need. That includes real-time cross-chain swaps, recurring buy schedules, portfolio tracking, token discovery and analysis, multi-chain deposit links with automatic stablecoin conversion, fiat funding via virtual accounts that accept bank transfers, Apple Pay, Venmo, and PayPal, and the ability to off-ramp back to traditional currencies from the terminal.
Multi-chain coverage at launch spans Ethereum, Solana, Base, Polygon, Arbitrum, Optimism, BNB Chain, Avalanche, TRON, and Bitcoin. Over 100 tokens are supported. Developers can also extend the platform with custom skills. The system is compatible with Claude, ChatGPT, Gemini, and Grok, and can be accessed via the CLI, a local Model Context Protocol server, or a web chat interface.
One detail that has caught the attention of developers in the agentic AI space is native x402 support. The x402 protocol, introduced by Coinbase in May 2025, revives the long-dormant HTTP 402 status code to enable machine-to-machine payments using stablecoins, with no API keys or subscriptions required. An agent simply pays for a resource or service at the time of access. MoonPay's inclusion of x402 compatibility positions MoonPay Agents within the emerging standard that Stripe, QuickNode (which extended x402 support across more than 80 chains), and a growing number of infrastructure providers have rallied around.
MoonPay Agents is not architected for one or two bots. The infrastructure is built to support thousands, eventually millions, of agents running concurrently across use cases that range from trading and portfolio management to gaming economies, commerce automation, and corporate treasury operations.
The Ledger Integration
MoonPay's solution was to bring Ledger into the loop. By integrating Ledger's Device Management Kit into the CLI wallet for MoonPay Agents, the company now allows every AI-generated transaction to be routed through a physical hardware device for approval. The agent constructs and proposes the transaction. The user confirms it on the Ledger. Private keys never touch the software layer at any point.
MoonPay says this makes the CLI wallet the first agent-focused wallet to support Ledger's secure signing through the Device Management Kit. Soto-Wright put the strategic framing plainly: "Autonomous agents will manage trillions in digital assets. But autonomy without security is reckless. We built MoonPay Agents with Ledger so intelligence can scale without surrendering control. The agent executes. The human stays in the loop."
Ledger's chief experience officer, Ian Rogers, acknowledged that the partnership reflects a real shift in what wallet infrastructure needs to support. "There is a new wave of CLI and agent-centric wallets emerging," he said, "and these will need Ledger security as a feature, too." It is a meaningful endorsement from a company whose entire value proposition is built on the premise that hardware is the only storage you can actually trust.
The model that results from the integration is structurally similar to two-factor authentication in traditional finance: the AI handles the analytical and execution work, but physical confirmation is required to release funds. Even a fully compromised software environment cannot move money without the physical Ledger device and its PIN.
For developers building agents that need to touch money, the practical implications of MoonPay Agents are fairly direct. The product abstracts away most of the hard parts: custody, key management, fiat connectivity, cross-chain routing, compliance. A single CLI install and a one-time user verification is genuinely all that stands between a developer and an agent that can fund itself, trade across chains, and off-ramp back to a bank account.
The ability to add custom skills also matters. MoonPay Agents ships with 54 tools across 17 categories, but the open extension model means developers can build on top of the existing toolkit rather than working around its edges. That kind of extensibility is usually what determines whether a platform becomes a default or a footnote.
What remains to be seen is how the ecosystem grows around it. MoonPay has the infrastructure and the user base. The question now is whether developers building the next generation of agentic applications pick MoonPay Agents as their default financial layer, or whether a competitor, or a collection of open standards, fills that space instead.
It is worth stepping back from the product details for a moment to consider what MoonPay is actually doing here. This is not a company adding AI features to an existing payments product. It is a payments company making a deliberate bet that the financial system is about to acquire a new class of participant, one that is not human, that will require infrastructure designed specifically for machine-speed, machine-scale capital movement, and that will need to be anchored to compliant fiat rails if it is ever going to interact with the broader economy.
That bet is not obviously wrong. Stablecoin volumes are growing at rates that would have seemed implausible even two years ago. Agent tokens and AI-driven trading systems are proliferating faster than most infrastructure providers anticipated. The convergence of AI and crypto, long discussed in the abstract, is becoming a concrete engineering problem that real companies are being paid to solve.
MoonPay's move is a claim that it has already built most of what that future requires, and that the work of this moment is connecting those existing rails to the autonomous systems that will run on them. It is an ambitious claim. The next 18 months will do a lot to determine whether it holds up.

Coinbase is not introducing AI agents to crypto. Those have been here for years.
What Coinbase is doing now is different. It is trying to formalize and secure that reality.
With the release of what it calls Agentic Wallets, Coinbase is offering wallet infrastructure built specifically for autonomous AI agents. Not dashboards with AI features. Not analytics copilots. Actual wallets engineered so software agents can hold and move funds in a way that is safer, cleaner, and more production ready than the duct taped setups many teams rely on today.
Erik Reppel, who leads engineering on the Coinbase Developer Platform, has been fairly direct about the problem they are solving.
Today, when developers say an agent “has a wallet,” that often means a private key is sitting somewhere it probably should not be. Maybe in a config file. Maybe in memory. Maybe loosely protected. If that agent gets manipulated, exploited, or simply misbehaves, the blast radius can be severe.
Reppel’s argument is that key isolation needs to be non negotiable. With Agentic Wallets, private keys are stored in secure execution environments, separated from the agent’s reasoning layer. The agent never directly touches raw key material. Instead, it interacts through controlled sessions with predefined permissions and limits.
He has described this architecture as orders of magnitude safer than letting an AI system operate with exposed keys.
That framing is important. Coinbase is not claiming to invent autonomous agents. It is trying to make them viable in production environments where security and compliance actually matter.
Two technical components sit at the core of this release: Base and x402.
Agentic Wallets are designed to run natively on Base, Coinbase’s Ethereum layer 2 network. Base offers lower fees and faster settlement compared to mainnet Ethereum, which makes it more practical for continuous automated activity. Bots and agents do not sleep. They monitor, adjust, and transact around the clock. Running that on a cheaper, faster chain is not a luxury, it is a necessity.
Then there is x402, Coinbase’s machine-to-machine payments protocol.
If that name sounds obscure, the idea is straightforward. x402 is built to allow services to pay other services directly onchain. It has already processed tens of millions of transactions in scenarios where APIs, compute layers, or other digital services require automated payment.
In the context of Agentic Wallets, x402 becomes the settlement layer for autonomous systems. An agent can pay for API access, purchase data feeds, cover inference costs, or settle fees with other services without a human approving every transaction. It is programmable, onchain, and designed for machines transacting with machines.
Put differently, Base provides the execution environment. x402 provides the payment rails. Agentic Wallets sit on top as the secure container that ties everything together.
It is worth saying clearly: AI driven trading is not new.
Quant desks, DeFi vaults, MEV bots, and arbitrage engines have been programmatically making trades for years. In many cases those systems are highly sophisticated. But the wallet layer underneath them has often been an afterthought. Keys get managed in inconsistent ways. Access control is custom built. Security depends heavily on the engineering discipline of each individual team.
What Coinbase is offering is a standardized wallet layer designed for autonomous operation from day one.
With Agentic Wallets, developers can:
That does not suddenly give agents new superpowers. They were already capable of executing trades, reallocating liquidity, and managing positions. What this does is reduce the fragility in how those systems are wired into capital.
For teams building serious onchain automation, that difference matters.
The safety architecture is arguably the most important part of this launch.
Prompt injection attacks, model manipulation, and logic exploits are not theoretical. If an agent is given broad financial authority and can be tricked into executing malicious instructions, the damage can be immediate and irreversible.
Coinbase’s model is to narrow the surface area.
Private keys live in secure enclaves. Agents operate through session credentials rather than raw key access. Developers can define how much value an agent can move and under what conditions. Transaction monitoring tools screen for high risk interactions before they are finalized.
None of this eliminates risk. Autonomous systems interacting with open financial networks will always carry some degree of unpredictability. But compared to the common practice of handing a bot a hot wallet and hoping for the best, this is a structural upgrade.
Zooming out, this fits into Coinbase’s broader strategy.
The company has been expanding its developer platform, pushing Base as a default settlement layer, and experimenting with tools that make onchain activity easier to embed into applications. Agentic Wallets extend that logic into the AI domain.
If AI systems continue to mediate financial activity, whether that is portfolio management, payments, or automated strategy execution, they will need infrastructure. Wallets are the choke point. Whoever controls that layer controls a meaningful slice of the stack.
Coinbase clearly wants to be that provider.
There are still regulatory and philosophical questions hanging over all of this. When an autonomous agent executes a trade or interacts with a protocol, who ultimately bears responsibility? The developer? The operator? The infrastructure provider? Those debates are just beginning.
But in practical terms, agents are already here. They are already trading. They are already moving markets.
Autonomous systems are currently participating in crypto. The wallet layer just needs to catch up.
Agentic Wallets are an attempt to do exactly that.