Artificial intelligence and machine learning applications
View all tags
A $12 trillion centralized AI empire on one side. A $12 billion decentralized challenger on the other. The gap is staggering — but it's closing faster than anyone predicted, and the blockchain AI market's projected leap from $6 billion to $50 billion by 2030 tells only part of the story.
Weekly open-source crypto commits have plummeted from 871,000 to 218,000 — a 75% collapse. Active blockchain developers dropped from 8,700 to 4,600, a 47% decline. Meanwhile, AI venture funding hit $211 billion in 2025 alone, absorbing talent at a pace the crypto industry has never faced. Yet hidden inside the wreckage is a paradox: the developers who stayed are more experienced, more productive, and shipping faster than ever before.
What if your AI assistant could not only analyze crypto markets but execute cross-chain swaps on your behalf — moving tokens from Ethereum to Solana in seconds, without you ever touching a bridge interface? That future arrived in February 2026 when deBridge launched the first open-source Model Context Protocol (MCP) server purpose-built for cross-chain DeFi execution.
The deBridge MCP server transforms AI coding assistants like Claude and Cursor from passive advisors into active cross-chain traders. It is part of a broader race — alongside Coinbase's Agentic Wallets, OKX's OnchainOS, and Bybit's AI Skills — to build the middleware layer that connects large language models to live blockchain liquidity. But deBridge's approach stands apart: instead of locking users into a single exchange's ecosystem, it routes trades across 26+ blockchains through a decentralized solver network with zero locked liquidity and full user custody.
This is not a speculative roadmap. It is production infrastructure, available today on GitHub, already integrated into developer workflows. And it signals a fundamental shift in how humans — and machines — will interact with decentralized finance.
What if you could own an AI agent the same way you own a cryptocurrency — transfer it, sell it, or watch it appreciate as it learns? On March 4, 2026, decentralized AI infrastructure protocol 0G launched AIverse on its Aristotle Mainnet, introducing what it calls the first "Web 4.0" marketplace. The platform turns AI agents into intelligent NFTs (iNFTs) — tokens that carry actual intelligence, memory, and capabilities rather than just a link to a JPEG.
Behind it all sits ERC-7857, a new Ethereum token standard purpose-built for tokenized intelligence. With over $300 million in ecosystem funding and 100+ partners including Chainlink, Google Cloud, and Samsung Next already building on 0G's infrastructure, iNFTs may represent the most ambitious attempt yet to make AI agents into tradeable economic actors.
Every AI agent needs a wallet. But who holds the keys?
On March 13, 2026, MoonPay answered that question by launching the first AI agent platform secured by a Ledger hardware signer — a move that forces every transaction through a physical device where private keys never touch the internet. In a market where 60–80% of global crypto trading volume is already AI-driven and autonomous agents manage billions in assets, MoonPay's bet is that the winning architecture isn't the one that moves fastest, but the one that humans still trust.
The crypto AI agent explosion of 2025–2026 created a paradox. Autonomous agents need wallet access to trade, bridge, stake, and pay for services. But wallet access means key access — and key access means trusting software with everything you own.
Before MoonPay's Ledger integration, the industry offered two imperfect options:
Full autonomy, zero security. Give the agent your private key or seed phrase. It can act instantly, but a single vulnerability — a prompt injection, a compromised dependency, a rogue API call — drains the wallet. In February 2026, supply chain attacks targeting dYdX through compromised npm and Python packages, linked to the Lazarus Group, demonstrated how real this threat is.
Full security, zero autonomy. Keep keys locked in cold storage and approve every transaction manually. Safe, but it defeats the purpose of autonomous agents entirely. You become the bottleneck in a system designed to operate at machine speed.
MoonPay's Ledger integration introduces a third path: autonomous strategy, human-verified execution. The AI agent handles research, portfolio analysis, swap routing, and trade construction. But every on-chain transaction must be physically confirmed on a Ledger device before it executes. The agent is the brain; the hardware wallet is the lock.
MoonPay Agents, initially released on February 24, 2026 as a command-line interface (CLI) tool, lets AI agents manage wallets, execute trades, and transact across multiple blockchains. The March 13 update adds native Ledger signer support, making it the first CLI wallet with this integration.
The technical flow is straightforward:
The critical security property: private keys are generated and stored inside the Ledger's secure element chip. They never leave the device, never touch the host computer's memory, and never enter the AI agent's execution environment. The agent can propose any action, but it cannot execute without human approval.
Available now in MoonPay CLI version 0.12.3 at moonpay.com/agents.
MoonPay's approach sits at one end of a security spectrum that the crypto industry is rapidly defining. Each major player has staked out a different position, and the tradeoffs reveal fundamentally different visions for how humans and AI agents should interact.
Coinbase launched its Agentic Wallets in February 2026, built on multi-party computation (MPC). Every action is signed by the agent using MPC and recorded on-chain on Ethereum or Base. Creators retain an emergency administrative key that can freeze or recover funds if malicious behavior is detected.
The model prioritizes programmability. Developers set spending limits, whitelisted contract interactions, and automated guardrails. The agent operates within defined boundaries without needing transaction-by-transaction human approval. It's closer to giving an employee a corporate card with spending limits than requiring a manager's signature on every purchase.
Tradeoff: Keys are managed in Coinbase's hosted infrastructure, not on a physical device the user controls. This is convenient for developers building autonomous systems but requires trusting Coinbase's custodial infrastructure.
At the opposite extreme, Coinbase's x402 protocol enables fully autonomous machine-to-machine payments with no human in the loop at all. Built directly into the HTTP layer, x402 lets AI agents pay for API calls, compute credits, and data access automatically using USDC on Base.
Alchemy integrated x402 in February 2026, creating a flow where an AI agent independently purchases compute credits and accesses blockchain data without any human intervention. The protocol has processed over 50 million transactions in testing, though daily real-world volume remains modest at roughly $28,000 — a sign that the infrastructure is ahead of adoption.
Tradeoff: Maximum speed and automation, but zero human oversight per transaction. Suitable for micropayments and API access, but risky for large trades or portfolio management.
MetaMask's approach uses session keys — temporary, scoped permissions that allow AI agents to perform specific actions while users retain full custody. Think of it as giving a valet your car key but programming it so it can only drive below 25 mph and can't open the trunk.
Tradeoff: More granular than MoonPay's all-or-nothing Ledger approval, but session keys are software-based, making them vulnerable to the same class of attacks that hardware wallets are designed to prevent.
MoonPay's Ledger integration occupies the maximum-security end of the spectrum. No transaction executes without a physical button press. This makes it the slowest option for high-frequency trading but the most resistant to software-based attacks, agent compromise, and unauthorized transactions.
As Ledger's chief experience officer noted: "There is a new wave of CLI and agent-centric wallets emerging, and these will need Ledger security as a feature, too."
The stakes are enormous. The agentic economy is projected to grow to $30 trillion by 2030, according to industry estimates. Microsoft reported in February 2026 that more than 80% of Fortune 500 companies now use active AI agents. In crypto specifically, over 550 AI agent projects exist with a combined market cap exceeding $4.3 billion, and AI quant funds reported average returns of 52% in 2025 while 84% of retail traders lost money.
The question isn't whether AI agents will manage crypto portfolios — they already do. The question is what security architecture becomes the institutional standard.
Three models are competing:
For retail users managing personal portfolios, hardware-in-the-loop may be ideal — the latency of pressing a button on a Ledger is irrelevant when you're making a few trades per day. For institutional quantitative strategies executing thousands of trades per second, it's a non-starter. For machine-to-machine micropayments, full autonomy is the only viable path.
The likely outcome isn't a single winner but a layered security stack. AI agents will use fully autonomous payments for sub-dollar API calls, MPC-secured wallets with spending limits for mid-range operations, and hardware-signed authorization for high-value transactions — the same way humans use tap-to-pay for coffee, a PIN for groceries, and a notary for real estate.
MoonPay's move signals that the AI agent infrastructure war is entering its security-differentiation phase. The first wave was about capability — can agents trade, bridge, and swap? That's solved. The second wave is about trust — can users and institutions deploy agents without risking catastrophic loss?
For developers building on-chain AI agents, the practical takeaways are:
Security architecture is now a product differentiator. Users will choose agent platforms based on how keys are managed, not just what strategies agents can execute.
Multi-tier security is inevitable. No single model serves all use cases. Build with pluggable key management that can support hardware signers, MPC, and session keys depending on transaction value and risk profile.
Regulatory scrutiny is coming. As AI agents manage larger portfolios, regulators will ask who is responsible when an agent makes unauthorized trades. Hardware-in-the-loop creates a clear audit trail: every transaction has a human-verified signature.
MoonPay's Ledger integration isn't a breakthrough in AI capability — the agents themselves don't get smarter. It's a breakthrough in the trust infrastructure that determines whether those agents get deployed at scale.
The crypto industry spent a decade learning that "not your keys, not your coins" is more than a slogan — it's an engineering requirement validated by exchange hacks, custodial failures, and billions in losses. Now, as AI agents ask for the same key access that centralized exchanges demanded, the industry faces the same question again: who holds the keys?
MoonPay's answer — a physical device that requires human confirmation for every transaction — is the most conservative possible response to the most important question in autonomous finance. In a market racing toward full automation, that conservatism might be exactly what institutions need to participate.
The agent economy will be built. The only question is whether it's built on a foundation of speed or a foundation of trust. MoonPay is betting that trust wins.
BlockEden.xyz provides enterprise-grade RPC and API infrastructure across Ethereum, Solana, Base, and 20+ blockchain networks — the foundational layer that AI agents depend on for reliable on-chain data and transaction submission. As autonomous agents demand secure, high-availability infrastructure, explore our API marketplace to build on foundations designed for the agentic era.
Within 90 days of each other in early 2026, every major payment platform on the planet launched its own AI agent payment protocol. Visa unveiled TAP. Google rallied 60 partners behind AP2. Coinbase shipped x402 with Cloudflare and Stripe backing. PayPal announced Agent Ready. The message was unmistakable: the companies that move trillions of dollars through the global economy are betting that, very soon, software — not humans — will initiate most of those transactions.
Gartner predicts that 40% of enterprise applications will embed task-specific AI agents by the end of 2026, up from less than 5% in 2025. The dedicated market for autonomous AI agent software is projected to reach $11.79 billion this year alone. And in the longer view, agentic AI could drive roughly 30% of enterprise application software revenue by 2035 — surpassing $450 billion. The race to become the TCP/IP of agent-initiated payments is not about next quarter's revenue. It is about who controls the rails for the next era of commerce.
What if you could trade not just what happens in the world, but what people think about it? In March 2026, Polymarket and Kaito AI launched exactly that — "Attention Markets," a new category of prediction markets where users wager on internet trends, brand popularity, and social sentiment rather than traditional real-world events. The partnership fuses Kaito's AI-quantified attention data with Polymarket's $21.5 billion prediction market infrastructure, creating tradeable instruments from something that has never been priced on-chain before: collective human attention.
The timing is no accident. It arrives just weeks after Kaito's flagship Yaps product was killed by X's API crackdown on InfoFi apps — and at a moment when prediction markets are projected to reach $1.3 trillion in annual volume by year-end.
In the past nine months, AI agents have completed 140 million payments totaling $43 million. Of those transactions, 98.6% settled in USDC — not because their developers love crypto, but because no other payment rail could do the job. That single statistic captures the most unexpected alliance in fintech: a technology community broadly skeptical of blockchain has quietly made stablecoins the default infrastructure for autonomous commerce.
In May 2021, Shiba Inu developers sent trillions of SHIB tokens to Vitalik Buterin's Ethereum wallet — unsolicited, uninvited, and intended purely as a marketing stunt. Nobody, least of all Buterin, expected what came next: those tokens surged past $1 billion in book value during the memecoin frenzy, and their liquidation quietly funded one of the most consequential — and controversial — pivots in AI policy advocacy history.
On March 14, 2026, a CoinDesk investigation revealed the full arc of this story. The Future of Life Institute (FLI), which received roughly half of Buterin's SHIB windfall, managed to liquidate approximately $500 million worth of the tokens — twenty to fifty times more than Buterin expected was even possible. That money has since been redirected from broad existential-risk research toward aggressive political lobbying on AI regulation, prompting the Ethereum co-founder to publicly distance himself from an organization he once supported.