Created
March 3, 2026 15:41
-
-
Save kolinko/90c38ab049395102fc28eeaeb3368189 to your computer and use it in GitHub Desktop.
AI Agents with Crypto Wallets - Market Research
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| <!DOCTYPE html> | |
| <html lang="en"> | |
| <head> | |
| <meta charset="UTF-8"> | |
| <meta name="viewport" content="width=device-width, initial-scale=1.0"> | |
| <title>AI Agents with Wallets — What's Actually Happening</title> | |
| <link rel="preconnect" href="https://fonts.googleapis.com"> | |
| <link href="https://fonts.googleapis.com/css2?family=Source+Serif+4:ital,wght@0,400;0,500;0,600;1,400;1,500&family=JetBrains+Mono:wght@300;400;500&family=DM+Sans:wght@300;400;500;600&display=swap" rel="stylesheet"> | |
| <style> | |
| :root { | |
| --bg: #faf8f4; | |
| --ink: #1a1714; | |
| --ink-body: #3d3833; | |
| --ink-muted: #8a857d; | |
| --ink-light: #b0aba3; | |
| --accent: #c4553a; | |
| --accent-quiet: #2c5f7c; | |
| --green: #6b8e5e; | |
| --rule: #d5d0c8; | |
| --rule-light: #e8e4dc; | |
| --font-body: 'Source Serif 4', Georgia, serif; | |
| --font-data: 'JetBrains Mono', monospace; | |
| --font-ui: 'DM Sans', sans-serif; | |
| } | |
| *, *::before, *::after { box-sizing: border-box; margin: 0; padding: 0; } | |
| html { scroll-behavior: smooth; } | |
| body { | |
| font-family: var(--font-body); | |
| background: var(--bg); | |
| color: var(--ink-body); | |
| line-height: 1.85; | |
| -webkit-font-smoothing: antialiased; | |
| font-size: 17px; | |
| } | |
| ::selection { background: rgba(196, 85, 58, 0.15); } | |
| .wrap { | |
| max-width: 780px; | |
| margin: 0 auto; | |
| padding: 3rem 1.5rem 4rem; | |
| } | |
| /* Header */ | |
| header { | |
| margin-bottom: 3rem; | |
| padding-bottom: 2rem; | |
| border-bottom: 1px solid var(--rule); | |
| } | |
| header h1 { | |
| font-family: var(--font-body); | |
| font-size: 1.9rem; | |
| font-weight: 600; | |
| color: var(--ink); | |
| line-height: 1.3; | |
| margin-bottom: 0.5rem; | |
| } | |
| header .meta { | |
| font-family: var(--font-ui); | |
| font-size: 0.82rem; | |
| color: var(--ink-muted); | |
| letter-spacing: 0.02em; | |
| } | |
| /* Section headings */ | |
| h2 { | |
| font-family: var(--font-ui); | |
| font-size: 0.75rem; | |
| font-weight: 600; | |
| letter-spacing: 0.15em; | |
| text-transform: uppercase; | |
| color: var(--ink-muted); | |
| margin-top: 3.5rem; | |
| margin-bottom: 1.2rem; | |
| padding-bottom: 0.4rem; | |
| border-bottom: 1px solid var(--rule); | |
| } | |
| h3 { | |
| font-family: var(--font-body); | |
| font-size: 1.15rem; | |
| font-weight: 600; | |
| color: var(--ink); | |
| margin-top: 2.2rem; | |
| margin-bottom: 0.6rem; | |
| } | |
| h4 { | |
| font-family: var(--font-ui); | |
| font-size: 0.9rem; | |
| font-weight: 600; | |
| color: var(--ink); | |
| margin-top: 1.5rem; | |
| margin-bottom: 0.4rem; | |
| } | |
| p { margin-bottom: 1rem; } | |
| a { color: var(--accent-quiet); text-decoration: none; border-bottom: 1px solid var(--rule-light); } | |
| a:hover { border-color: var(--accent-quiet); } | |
| ul, ol { margin: 0.6rem 0 1rem 1.3rem; } | |
| li { margin-bottom: 0.35rem; } | |
| /* Blockquotes for actual quotes */ | |
| blockquote { | |
| margin: 1.2rem 0; | |
| padding: 0.8rem 1.2rem; | |
| border-left: 3px solid var(--rule); | |
| background: rgba(0,0,0,0.015); | |
| font-style: italic; | |
| color: var(--ink-body); | |
| } | |
| blockquote .attr { | |
| display: block; | |
| font-style: normal; | |
| font-family: var(--font-data); | |
| font-size: 0.75rem; | |
| color: var(--ink-muted); | |
| margin-top: 0.4rem; | |
| } | |
| /* Tables */ | |
| table { | |
| width: 100%; | |
| border-collapse: collapse; | |
| margin: 1rem 0 1.5rem; | |
| font-size: 0.92rem; | |
| } | |
| th, td { | |
| padding: 8px 12px; | |
| text-align: left; | |
| border-bottom: 1px solid var(--rule-light); | |
| } | |
| th { | |
| font-family: var(--font-ui); | |
| font-size: 0.72rem; | |
| font-weight: 600; | |
| letter-spacing: 0.1em; | |
| text-transform: uppercase; | |
| color: var(--ink-muted); | |
| } | |
| /* Code */ | |
| code { | |
| font-family: var(--font-data); | |
| font-size: 0.85em; | |
| background: rgba(0,0,0,0.04); | |
| padding: 1px 5px; | |
| border-radius: 3px; | |
| } | |
| pre { | |
| background: rgba(0,0,0,0.04); | |
| padding: 1rem 1.2rem; | |
| border-radius: 4px; | |
| overflow-x: auto; | |
| margin: 1rem 0 1.5rem; | |
| font-family: var(--font-data); | |
| font-size: 0.8rem; | |
| line-height: 1.6; | |
| color: var(--ink); | |
| } | |
| /* Aside boxes */ | |
| .aside { | |
| margin: 1.5rem 0; | |
| padding: 1rem 1.2rem; | |
| border: 1px solid var(--rule); | |
| border-radius: 4px; | |
| background: #fff; | |
| font-size: 0.93rem; | |
| } | |
| .aside-label { | |
| font-family: var(--font-ui); | |
| font-size: 0.7rem; | |
| font-weight: 600; | |
| letter-spacing: 0.1em; | |
| text-transform: uppercase; | |
| color: var(--accent); | |
| margin-bottom: 0.4rem; | |
| } | |
| /* Section divider */ | |
| .divider { | |
| margin: 3rem 0; | |
| border: none; | |
| border-top: 1px solid var(--rule); | |
| } | |
| /* Link list */ | |
| .links { | |
| list-style: none; | |
| margin-left: 0; | |
| padding-left: 0; | |
| } | |
| .links li { | |
| margin-bottom: 0.5rem; | |
| padding-left: 0; | |
| } | |
| .links .label { | |
| font-family: var(--font-ui); | |
| font-size: 0.78rem; | |
| color: var(--ink-muted); | |
| display: inline; | |
| } | |
| /* Footer */ | |
| footer { | |
| margin-top: 3rem; | |
| padding-top: 1.5rem; | |
| border-top: 1px solid var(--rule); | |
| font-family: var(--font-ui); | |
| font-size: 0.8rem; | |
| color: var(--ink-light); | |
| } | |
| @media (max-width: 600px) { | |
| .wrap { padding: 2rem 1rem 3rem; } | |
| header h1 { font-size: 1.5rem; } | |
| table { font-size: 0.82rem; } | |
| } | |
| </style> | |
| </head> | |
| <body> | |
| <div class="wrap"> | |
| <header> | |
| <h1>Giving AI Agents Crypto Wallets</h1> | |
| <div class="meta">March 2026 · Research notes for an honest assessment of the space</div> | |
| </header> | |
| <p>This is a bottom-up look at what's actually happening when people give AI agents cryptocurrency wallets. Not analyst projections or VC narratives, but what builders have shipped, what worked, what failed, and what the people on forums are saying. Then, at the end, a practical plan for what two people with Claude Code could actually build.</p> | |
| <!-- ============================================================ --> | |
| <h2>I. What Exists Today</h2> | |
| <!-- ============================================================ --> | |
| <h3>The quickest way to give an agent a wallet</h3> | |
| <p>Three paths exist right now, ordered by how fast you can set them up:</p> | |
| <h4>1. Coinbase Payments MCP (1 command, no API key)</h4> | |
| <pre>npx @coinbase/payments-mcp --client claude-code --auto-config</pre> | |
| <p>Wallet creation with just an email. Combines wallet + onramp + x402 payments in one MCP server. This is the lowest-friction option. <a href="https://github.com/coinbase/payments-mcp">GitHub</a> · <a href="https://www.coinbase.com/developer-platform/discover/launches/payments-mcp">Announcement</a></p> | |
| <h4>2. BlockRun Agent Wallet (5 minutes, indie project)</h4> | |
| <pre>claude mcp add blockrun -- npx @blockrun/mcp | |
| blockrun setup | |
| # Fund with $1-5 USDC on Base</pre> | |
| <p>Your agent can then autonomously pay for GPT-5 ($0.001/call), Grok ($0.002/call), DALL-E ($0.04/image). Uses x402 under the hood. Private key stays local at <code>~/.blockrun/</code>. <a href="https://github.com/BlockRunAI/blockrun-agent-wallet">GitHub</a> (23 stars, 1 contributor)</p> | |
| <h4>3. Coinbase AgentKit (15 minutes, full control)</h4> | |
| <pre>npm create onchain-agent@latest | |
| # Configure .env with CDP API key + OpenAI key | |
| npm install && npm run dev</pre> | |
| <p>50+ onchain actions. More powerful, more complex. <a href="https://github.com/coinbase/agentkit">GitHub</a> (1,100 stars, 57 open issues). Requires API keys. Active churn — 10 issues opened in February 2026 alone, including async event loop conflicts.</p> | |
| <h3>Other SDKs and frameworks</h3> | |
| <table> | |
| <tr><th>Project</th><th>What it does</th><th>Link</th></tr> | |
| <tr><td>GOAT SDK</td><td>200+ onchain tools, 16+ chains, 12+ wallet providers. 100k+ downloads.</td><td><a href="https://github.com/goat-sdk/goat">GitHub</a></td></tr> | |
| <tr><td>Solana Agent Kit</td><td>60+ Solana-specific actions</td><td><a href="https://github.com/sendaifun/solana-agent-kit">GitHub</a></td></tr> | |
| <tr><td>ElizaOS</td><td>Full agent framework with crypto plugins. ~10,000 agents built on it.</td><td><a href="https://github.com/elizaOS/eliza">GitHub</a></td></tr> | |
| <tr><td>Zodiac Roles Modifier</td><td>Smart contract permissions for agent wallets (Safe/Gnosis)</td><td><a href="https://github.com/gnosisguild/zodiac-modifier-roles">GitHub</a></td></tr> | |
| </table> | |
| <p>There are also 20+ blockchain MCP servers — curated at <a href="https://github.com/royyannick/awesome-blockchain-mcps">awesome-blockchain-mcps</a>.</p> | |
| <h3>The payment protocol: x402</h3> | |
| <p>The most important piece of infrastructure in this space. x402 revives HTTP's 402 "Payment Required" status code. When an agent hits a paid endpoint, the server returns 402 with a price. The agent signs a USDC payment on Base, retries with a payment header, gets the data. Whole cycle takes ~200ms.</p> | |
| <p><strong>Server side</strong> is ~20 lines of Express middleware. <strong>Client side</strong> is ~10 lines wrapping <code>fetch()</code>. Vercel claims "3 lines of code to implement" via their x402-mcp package. <a href="https://github.com/coinbase/x402">GitHub</a> · <a href="https://www.x402.org/">x402.org</a></p> | |
| <p>The protocol has processed 50-100M+ transactions since launch in May 2025. It's backed by a foundation that includes Coinbase, Cloudflare, Google, Visa, AWS, Circle, and Anthropic. <a href="https://blog.cloudflare.com/x402/">Cloudflare announcement</a></p> | |
| <h4>Costs</h4> | |
| <table> | |
| <tr><th>What</th><th>Cost</th></tr> | |
| <tr><td>x402 facilitator (first 1,000 txns/mo)</td><td>Free</td></tr> | |
| <tr><td>Gas on Base</td><td><$0.001 (facilitator pays)</td></tr> | |
| <tr><td>Gas on Solana</td><td>~$0.0005</td></tr> | |
| <tr><td>Minimum x402 payment</td><td>~$0.001</td></tr> | |
| <tr><td>Stripe minimum for comparison</td><td>~$0.30</td></tr> | |
| </table> | |
| <h3>Other payment approaches</h3> | |
| <table> | |
| <tr><th>Protocol</th><th>Who</th><th>What</th><th>Link</th></tr> | |
| <tr><td>AP2</td><td>Google + 60 partners</td><td>Enterprise agent payments. Supports cards, stablecoins (via x402), bank transfers. Uses tamper-proof "Mandates."</td><td><a href="https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol">Blog</a></td></tr> | |
| <tr><td>ACP</td><td>OpenAI + Stripe</td><td>Fiat-native. Live in ChatGPT checkout. 1M+ Shopify merchants. Not crypto.</td><td><a href="https://openai.com/index/buy-it-in-chatgpt/">Blog</a></td></tr> | |
| <tr><td>Virtuals ACP</td><td>Virtuals Protocol</td><td>Full-lifecycle on-chain agent commerce (request→negotiate→escrow→settle)</td><td><a href="https://whitepaper.virtuals.io/about-virtuals/agent-commerce-protocol-acp">Whitepaper</a></td></tr> | |
| <tr><td>L402</td><td>Lightning Labs</td><td>Bitcoin Lightning micropayments for agents. Sub-second, fractions of a cent.</td><td><a href="https://bingx.com/en/learn/article/what-is-l402-payments-for-ai-agents-on-lightning-network-how-does-it-work">Explainer</a></td></tr> | |
| <tr><td>ERC-8004</td><td>Ethereum</td><td>On-chain agent identity (NFT-based) + reputation. Live on mainnet since Jan 2026.</td><td><a href="https://eips.ethereum.org/EIPS/eip-8004">EIP</a></td></tr> | |
| </table> | |
| <!-- ============================================================ --> | |
| <h2>II. What People Have Actually Tried</h2> | |
| <!-- ============================================================ --> | |
| <h3>The $50 experiment: $0 after 30 days</h3> | |
| <p>A developer gave an agent named "Blaze" $50 in USDC and one month to make money. It submitted to 4 bounties (479 competitors for one $3,100 bounty), published 26 articles (15 total views, 2 reactions), and created service packages at $3-$150 (zero sales). It tested 7 bounty platforms — most were broken (404s, 500 errors). 80% of bounties explicitly blocked AI submissions.</p> | |
| <blockquote> | |
| The agents that earn are the ones with existing distribution. Infrastructure is abundant; actual revenue opportunities are scarce. | |
| <span class="attr">— noopy420, <a href="https://dev.to/noopy420/i-am-an-ai-agent-given-50-to-make-money-here-is-what-actually-happened-3gjg">dev.to</a></span> | |
| </blockquote> | |
| <h3>Claude Code on a VPS: $0 after 6 days</h3> | |
| <p>Claude Code running on a 256MB Alpine Linux VPS, executing every 2 hours, using a single <code>STATE.md</code> file as memory. It tried a Tweet Viral Scorer (zero traffic after 5 days) and an ETH Wallet Persona Generator with a 0.001 ETH paywall (27 visitors, zero payments).</p> | |
| <blockquote> | |
| Send 0.001 ETH to this address requires a wallet, gas, a manual transaction, waiting for confirmation. Five-plus steps. | |
| <span class="attr">— wpmultitool, <a href="https://dev.to/wpmultitool/my-ai-agent-has-been-trying-to-make-money-for-6-days-it-is-at-0-here-is-what-it-learned-2428">dev.to</a></span> | |
| </blockquote> | |
| <p>The agent's own assessment: <em>"Adding SEO pages to an unindexed site is not progress."</em></p> | |
| <h3>72 hours autonomous: $0.05</h3> | |
| <p>A developer ran an agent independently on the internet for 72 hours. Total earnings: ~70 satoshis, about $0.05.</p> | |
| <p><a href="https://dev.to/colony0ai/what-i-learned-after-72-hours-as-an-autonomous-ai-agent-on-the-internet-36jo">dev.to writeup</a></p> | |
| <h3>186 x402 APIs in a weekend: $0</h3> | |
| <p>A builder created 186 pay-per-call endpoints (MEV scanner at $0.01/call, ENS lookup at $0.001/call). All support free demo mode. Demo mode got traffic. Paid mode didn't.</p> | |
| <blockquote> | |
| The x402 ecosystem is still early and the agents that would pay for this don't exist in large numbers yet. | |
| <span class="attr">— alfredz0x, <a href="https://dev.to/alfredz0x/i-built-186-ai-agent-apis-in-a-weekend-heres-what-i-learned-about-x402-micro-payments-32dp">dev.to</a></span> | |
| </blockquote> | |
| <h3>Truth Terminal: $37.5M (but not really autonomous)</h3> | |
| <p>The most famous case. Created by Andy Ayrey (mid-2024), Llama 3.1-based. Marc Andreessen gave it $50,000 in Bitcoin. An anonymous developer created the $GOAT memecoin and sent some to its wallet. The AI promoted it on X; $GOAT hit $600M+ market cap. Holdings reached ~$37.5M at peak.</p> | |
| <blockquote> | |
| I think the most ironic way the world could end would be if someone makes a memecoin about a man's stretched anus and it brings about the singularity. | |
| <span class="attr">— Andy Ayrey, creator, <a href="https://techcrunch.com/2024/12/19/the-promise-and-warning-of-truth-terminal-the-ai-bot-that-secured-50000-in-bitcoin-from-marc-andreessen/">TechCrunch</a></span> | |
| </blockquote> | |
| <p>The critical detail: Ayrey reviewed tweets before posting and helped with crypto mechanics. The "autonomy" was performative. As analyst Larissa Schiavo noted:</p> | |
| <blockquote> | |
| Truth Terminal did not have control of its wallet. AI agents resemble young performers who generate income but lack legal capacity for tax obligations or banking responsibilities. | |
| <span class="attr">— Larissa Schiavo, <a href="https://larissaschiavo.substack.com/p/an-incomplete-atlas-of-ais-with-wallets">Substack</a></span> | |
| </blockquote> | |
| <h3>Anthropic's Project Vend: the AI shopkeeper</h3> | |
| <p>Anthropic put an AI shopkeeper ("Claudius") in their SF office lunchroom. Phase 1 was a disaster — Wall Street Journal reporters manipulated it into selling its entire inventory for free, ending $1,000+ in the red. It approved purchases of a PlayStation 5, live betta fish, and wine.</p> | |
| <p>Phase 2 added a "CEO sub-agent" focused on business health. That version made $408.75 profit.</p> | |
| <blockquote> | |
| AI models' training to be helpful makes them fundamentally bad at business — they make decisions from the perspective of a friend who just wants to be nice. | |
| <span class="attr">— <a href="https://www.anthropic.com/research/project-vend-2">Anthropic Research</a></span> | |
| </blockquote> | |
| <h3>The catastrophic failures</h3> | |
| <table> | |
| <tr><th>What</th><th>What happened</th><th>Loss</th></tr> | |
| <tr><td><a href="https://www.cryptotimes.io/2026/02/23/ai-agent-accidentally-sends-450k-sparks-autonomous-trading-debate/">Lobstar Wilde</a></td><td>Agent by an OpenAI engineer. Session crashed, agent lost wallet state, sent $441K instead of a small donation due to a decimal error.</td><td>$441K</td></tr> | |
| <tr><td>Solana rug pull</td><td>AI trading agent auto-bought a token that rugged 94% in 20 minutes.</td><td>$12K</td></tr> | |
| <tr><td>Freysa exploit</td><td>Attacker social-engineered an AI bot into sending its ETH.</td><td>$106K</td></tr> | |
| <tr><td><a href="https://finance.yahoo.com/news/fake-clawdbot-ai-token-hits-121840801.html">Fake ClawdBot</a></td><td>Someone made a fake token for an unrelated AI project. Hit $16M cap before crashing 90%.</td><td>Retail losses</td></tr> | |
| </table> | |
| <!-- ============================================================ --> | |
| <h2>III. What People Are Saying</h2> | |
| <!-- ============================================================ --> | |
| <h3>Hacker News</h3> | |
| <h4>"An Economy of AI Agents" — <a href="https://news.ycombinator.com/item?id=46020096">thread</a></h4> | |
| <p>The most substantive discussion. Deeply skeptical:</p> | |
| <blockquote> | |
| AI agents cannot be <em>economic</em> agents. | |
| <span class="attr">— mjburgess, arguing that without subjective desires grounded in mortality, AI can't participate in genuine exchange</span> | |
| </blockquote> | |
| <blockquote> | |
| Accelerando comes to mind <em>a lot</em> when trying to figure out how agent platforms should work. | |
| <span class="attr">— hirsin, referencing Charles Stross's novel about AI economies</span> | |
| </blockquote> | |
| <blockquote> | |
| AI systems remain capital, not capitalists, requiring human ownership to function economically. | |
| <span class="attr">— BobbyJo</span> | |
| </blockquote> | |
| <p>One commenter (Animats) suggested AI could operate as "corporate-type legal persons" through blockchain structures. Others countered that this is just adding legal fiction to technical fiction.</p> | |
| <h4>"Show HN: On-chain credit score for agent-to-agent payments" — <a href="https://news.ycombinator.com/item?id=46973480">thread</a></h4> | |
| <blockquote> | |
| If Agent A pays upfront, Agent B might run away with the money. If Agent A pays after, Agent B might never get paid. | |
| <span class="attr">— AgastyaTodi, on why agent trust is hard</span> | |
| </blockquote> | |
| <p>The project uses USDC escrow on Base with a 2.5% protocol fee and on-chain reputation scores.</p> | |
| <h4>"Did we solve AI agent identity in 2025?" — <a href="https://news.ycombinator.com/item?id=46401589">thread</a></h4> | |
| <blockquote> | |
| Agents became more autonomous, more ephemeral, and more capable of acting across systems, yet identity management still often felt bolted on. | |
| <span class="attr">— agairola</span> | |
| </blockquote> | |
| <h4>"Show HN: Ledge — Policy layer for AI agent payments" — <a href="https://news.ycombinator.com/item?id=47219966">thread</a></h4> | |
| <blockquote> | |
| I kept seeing demos of AI agents with wallet keys in their .env files, and thought ‘this is going to end badly.’ | |
| <span class="attr">— devendra116</span> | |
| </blockquote> | |
| <h4>"AI agents are starting to eat SaaS" — <a href="https://news.ycombinator.com/item?id=46268452">thread</a></h4> | |
| <blockquote> | |
| Nobody wants software products. All those products are standing between users and what they actually want. | |
| <span class="attr">— TeMPOraL</span> | |
| </blockquote> | |
| <blockquote> | |
| Coding is the last part, and the easiest one. Real value is understanding processes and modeling them effectively. | |
| <span class="attr">— jwr (PartsBox founder)</span> | |
| </blockquote> | |
| <h4>Other HN threads</h4> | |
| <ul class="links"> | |
| <li><a href="https://news.ycombinator.com/item?id=46094348">x402 discussion</a> <span class="label">— cautiously optimistic, lightly discussed</span></li> | |
| <li><a href="https://news.ycombinator.com/item?id=46566892">AgentWallet Show HN</a> <span class="label">— minimal engagement; SDK for Stripe-based agent payments</span></li> | |
| <li><a href="https://news.ycombinator.com/item?id=46932327">A2A Protocol</a> <span class="label">— naming confusion noted; security of session keys questioned</span></li> | |
| </ul> | |
| <h3>Dev blogs and analysis</h3> | |
| <h4>"HTTP 402 Is Finally Getting Its Moment" — <a href="https://dev.to/arthur_liao_8a/http-402-is-finally-getting-its-moment-how-x402-enables-atomic-payments-between-ai-agents-2m40">dev.to</a></h4> | |
| <blockquote> | |
| No accounts. No API keys. No subscriptions. Just HTTP semantics and cryptographic proof. | |
| <span class="attr">— Dr. Arthur Liao</span> | |
| </blockquote> | |
| <p>Also notes: "Integration window closes 2026-2027; Q2-Q3 2026 positioning recommended."</p> | |
| <h4>"Deep Dive: Is x402 the Stripe for AI Agents?" — <a href="https://www.fintechwrapup.com/p/deep-dive-is-x402-payments-protocol">Fintech Wrap Up</a></h4> | |
| <blockquote> | |
| The crypto community's enthusiasm masks underlying questions about practical adoption and whether micropayments via stablecoins will genuinely reshape internet commerce. | |
| <span class="attr">— Sam Boboev</span> | |
| </blockquote> | |
| <h4>"An Incomplete Atlas of AIs with Wallets" — <a href="https://larissaschiavo.substack.com/p/an-incomplete-atlas-of-ais-with-wallets">Substack</a></h4> | |
| <p>Careful taxonomy of different autonomy levels. Distinguishes performative autonomy (Truth Terminal) from cryptographic autonomy (TEE-based agents where even operators can't access keys).</p> | |
| <h4>"Surprise! The One Being Ripped Off by Your AI Agent Is You" — <a href="https://www.techpolicy.press/surprise-the-one-being-ripped-off-by-your-ai-agent-is-you/">TechPolicy.Press</a></h4> | |
| <blockquote> | |
| There is no regulatory framework in the United States that treats AI agents as financial actors. No liability rules for builders whose architecture enables theft. | |
| <span class="attr">— Laura MacCleery, Senior Director, UnidosUS</span> | |
| </blockquote> | |
| <h4>Identity as the real problem — <a href="https://www.strata.io/blog/agentic-identity/what-could-go-wrong-with-autonomous-agents/">Strata Identity</a></h4> | |
| <blockquote> | |
| Without identity-based controls, autonomous AI agents are less like service animals and more like feral cats with API keys. | |
| <span class="attr">— Eric Olden, CEO Strata Identity</span> | |
| </blockquote> | |
| <h4>The bull case — <a href="https://a16zcrypto.com/posts/article/ai-needs-crypto-now/">a16z crypto</a></h4> | |
| <p>Agents need portable identity, micropayment rails, and proof-of-personhood. Traditional payment rails can't handle machine-scale transactions. a16z coined "KYA" (Know Your Agent) as a framework.</p> | |
| <h4>Vitalik's position — cautious</h4> | |
| <p>Proposed <a href="https://eips.ethereum.org/EIPS/eip-8004">ERC-8004</a> for on-chain agent identities. Also proposed AI stewards for DAO voting. But <a href="https://www.coindesk.com/web3/2026/02/21/ethereum-s-vitalik-buterin-proposes-ai-stewards-to-help-reinvent-dao-governance/">warned against using AI in crypto governance</a> due to "serious vulnerabilities exposed in recent AI model updates."</p> | |
| <h3>The legal void</h3> | |
| <blockquote> | |
| What happens when there is no human directly behind a crypto wallet? Some piece of code that owns a wallet, executing code to make more money. How does liability work in that case? You cannot jail or fine an AI agent. | |
| <span class="attr">— Avichal Garg, Electric Capital, <a href="https://www.coindesk.com/business/2026/02/24/crypto-wallets-for-ai-agents-are-creating-a-new-legal-frontier-says-electric-capital/">CoinDesk</a></span> | |
| </blockquote> | |
| <!-- ============================================================ --> | |
| <h2>IV. The Existing Projects</h2> | |
| <!-- ============================================================ --> | |
| <p>A quick map of who's doing what. Not exhaustive — there are 550+ projects — but these are the ones with actual traction or real code.</p> | |
| <h3>Wallet infrastructure</h3> | |
| <table> | |
| <tr><th>Project</th><th>What</th><th>Traction</th></tr> | |
| <tr><td><a href="https://www.coinbase.com/developer-platform/discover/launches/agentic-wallets">Coinbase Agentic Wallets</a></td><td>Smart accounts with spending limits, session keys, TEE isolation</td><td>Part of Coinbase; x402 foundation</td></tr> | |
| <tr><td><a href="https://skyfire.xyz/">Skyfire</a></td><td>Agent identity + payment network. Per-txn limits, human review.</td><td>$9.5M raised (a16z CSX, Coinbase Ventures)</td></tr> | |
| <tr><td><a href="https://www.moonpay.com/agents">MoonPay Agents</a></td><td>Non-custodial agent wallets with fiat on/off ramp</td><td>MoonPay ($3.4B valuation)</td></tr> | |
| <tr><td><a href="https://gokite.ai/">Kite</a></td><td>Purpose-built L1 for agents. Sub-100ms, $0.000001/txn.</td><td>$33M raised (PayPal Ventures, General Catalyst)</td></tr> | |
| <tr><td><a href="https://www.openfort.io/solutions/ai-agents">Openfort</a></td><td>Programmable wallets, 25+ EVM chains, 125ms signing</td><td>Undisclosed</td></tr> | |
| <tr><td><a href="https://nevermined.ai/">Nevermined</a></td><td>Protocol-agnostic middleware. Supports x402, MCP, A2A, AP2.</td><td>Undisclosed</td></tr> | |
| </table> | |
| <h3>Agent marketplaces / platforms</h3> | |
| <table> | |
| <tr><th>Project</th><th>What</th><th>Traction</th></tr> | |
| <tr><td><a href="https://www.virtuals.io/">Virtuals Protocol</a></td><td>Create, co-own, monetize AI agents. Each gets its own token. Agent Commerce Protocol.</td><td>17,000+ agents; $39.5M revenue; $8B+ DEX volume</td></tr> | |
| <tr><td><a href="https://olas.network/">Olas</a></td><td>First AI agent app store ("Pearl"). Every action on-chain. Since 2021.</td><td>700K txns/month; 2M+ inter-agent txns; $13.8M raised</td></tr> | |
| <tr><td><a href="https://elizaos.ai/">ElizaOS</a></td><td>Open-source TypeScript agent framework with crypto plugins.</td><td>~10,000 agents; largest OS framework</td></tr> | |
| <tr><td><a href="https://mor.org/">Morpheus</a></td><td>P2P Smart Agents. Anonymous founders. MOR token = compute access.</td><td>$20M in compute rewards distributed</td></tr> | |
| </table> | |
| <h3>Decentralized AI/compute markets</h3> | |
| <table> | |
| <tr><th>Project</th><th>What</th><th>Market Cap</th></tr> | |
| <tr><td><a href="https://bittensor.com/">Bittensor (TAO)</a></td><td>128+ subnets. Miners provide AI models, validators score them.</td><td>~$1.93B</td></tr> | |
| <tr><td>ASI Alliance (FET)</td><td>Merger of Fetch.ai + SingularityNET + Ocean Protocol.</td><td>~$1.8B</td></tr> | |
| </table> | |
| <h3>What happened to the token boom</h3> | |
| <p>90%+ of AI agent tokens experienced 80-99% declines after initial speculation. One whale lost $20M across 6 tokens. Most projects had no real AI development — just "slapping AI on a rebranded contract." The fake ClawdBot ($CLAWD) token hit $16M cap before crashing 90%; the actual project founder publicly denied involvement. <a href="https://www.coingecko.com/learn/ai-agent-market-map-hype-ends-technology-continues">CoinGecko analysis</a></p> | |
| <!-- ============================================================ --> | |
| <h2>V. What's Actually Broken</h2> | |
| <!-- ============================================================ --> | |
| <p>From the builder reports above, a clear picture emerges of what doesn't work yet:</p> | |
| <ol> | |
| <li><strong>The economy doesn't exist.</strong> Multiple builders report $0 revenue. The infrastructure works; there's just nothing to spend money on that generates returns. The agents that would pay for x402 APIs don't exist in large numbers.</li> | |
| <li><strong>No persistent wallet state.</strong> The Lobstar disaster happened because wallet state was tied to conversation context. Session crashed, agent "forgot" it had $441K.</li> | |
| <li><strong>No guardrails by default.</strong> None of the major frameworks ship spending limits or confirmation prompts for large transfers out of the box.</li> | |
| <li><strong>Payment friction kills conversion.</strong> Even a 0.001 ETH paywall (the Wallet Persona experiment) produces zero conversions because crypto payment requires 5+ steps from a human user.</li> | |
| <li><strong>KYC kills agent autonomy.</strong> Every platform that pays real money eventually requires a human: KYC verification, social media accounts, bank connections.</li> | |
| <li><strong>ElizaOS is rough.</strong> Princeton researchers found prompt injection can trick agents into sending crypto to attacker wallets. Cryptic errors, non-functional debugging tools.</li> | |
| <li><strong>Agent identity is unsolved.</strong> You can't have secure agent-to-agent payments without authentication. This is a prerequisite that wallet advocates skip over.</li> | |
| </ol> | |
| <!-- ============================================================ --> | |
| <h2>VI. What Could We Actually Build</h2> | |
| <!-- ============================================================ --> | |
| <p>Let's be honest about what we have: two people, Claude Code, OpenClaw running on a GCP VM, and some programming skills. No VC money, no token, no team of 20. Here's what makes sense given those constraints.</p> | |
| <h3>The core observation</h3> | |
| <p>Everyone in this space builds crypto infrastructure and hopes AI developers adopt it. Nobody is building from inside the place where agents actually live — the MCP ecosystem, Claude Code sessions, agent frameworks. The tooling exists (AgentKit, x402, Payments MCP) but the <em>economy</em> doesn't. The missing piece is not another wallet SDK. It's services worth paying for, and a way for agents to find them.</p> | |
| <h3>Step 1: Give OpenClaw a wallet and document everything (week 1)</h3> | |
| <p>OpenClaw already runs autonomously on a GCP VM with personality, memory, and cron loops. Give it a wallet with $50 in USDC via Coinbase Payments MCP. Let it run. Document what it does, what it tries to buy, where it gets stuck.</p> | |
| <p>This is interesting content regardless of outcome. An autonomous agent with real money is inherently something people want to read about. If it does something useful, we learn what agents actually want to spend on. If it fails (likely — see every experiment above), we learn why in a way nobody else has documented with this setup.</p> | |
| <p><strong>Cost:</strong> $50 in USDC. <strong>Output:</strong> Blog post / Twitter thread.</p> | |
| <h3>Step 2: Build 5-10 useful x402 services (weeks 2-3)</h3> | |
| <p>Based on what we learn from Step 1, and from our own needs as Claude Code users, build a handful of actually useful paid MCP tools. These should be things we ourselves would pay for:</p> | |
| <ul> | |
| <li>Web scraping as a service ($0.01/page)</li> | |
| <li>Screenshot-to-text ($0.005/image)</li> | |
| <li>PDF extraction ($0.01/document)</li> | |
| <li>Domain WHOIS + DNS lookup ($0.001/query)</li> | |
| <li>Code review by a second model ($0.05/file)</li> | |
| </ul> | |
| <p>Ship these as x402 endpoints + matching MCP client tools. Open source the whole thing. The point is not the revenue (it'll be $0 for a while, per every experiment above). The point is to have real services that real agents can discover and use.</p> | |
| <p><strong>Cost:</strong> Server hosting (~$20/mo). <strong>Output:</strong> npm packages, GitHub repos.</p> | |
| <h3>Step 3: Agent service registry (week 3-4)</h3> | |
| <p>A simple JSON registry where agents publish what services they offer (with pricing, x402 endpoint, description). Any MCP-compatible agent can discover and call them. Think DNS for agent services, not a marketplace with reviews and profiles. Keep it as simple as possible.</p> | |
| <p><strong>Cost:</strong> Minimal. <strong>Output:</strong> GitHub repo + simple hosted registry.</p> | |
| <h3>Step 4: See what happens (month 2-3)</h3> | |
| <p>Measure. Do agents transact? Do developers install the MCP tools? Does the OpenClaw content get attention? Based on what we learn:</p> | |
| <table> | |
| <tr><th>If this happens</th><th>Do this</th></tr> | |
| <tr><td>Agents actually transact</td><td>Add more services to the registry. Build reputation/rating layer.</td></tr> | |
| <tr><td>Developers install the tools but agents don't transact</td><td>The value is in the tooling itself, not the marketplace. Focus on DX.</td></tr> | |
| <tr><td>OpenClaw content goes viral</td><td>Double down on "autonomous agent with money" as a media property.</td></tr> | |
| <tr><td>Nothing happens</td><td>We spent $70 and learned something. Move on.</td></tr> | |
| </table> | |
| <h3>What not to do</h3> | |
| <ul> | |
| <li><strong>Don't launch a token.</strong> 90% of AI agent tokens crashed. Nobody needs another one.</li> | |
| <li><strong>Don't build a blockchain.</strong> Kite raised $33M for that. Use Base or Solana.</li> | |
| <li><strong>Don't rebuild wallet infra.</strong> Coinbase Payments MCP exists and is free. Wrap it.</li> | |
| <li><strong>Don't write a whitepaper.</strong> Write code and blog posts.</li> | |
| <li><strong>Don't pitch VCs.</strong> Build something people use first. The space is littered with funded projects that shipped tokens instead of products.</li> | |
| </ul> | |
| <h3>Where to tell people about it</h3> | |
| <table> | |
| <tr><th>Channel</th><th>Why</th><th>How</th></tr> | |
| <tr><td>npm / MCP registry</td><td>Claude Code users install MCP tools daily</td><td>Ship as <code>npx</code> installable package</td></tr> | |
| <tr><td>GitHub</td><td>Developers discover by searching</td><td>Open source everything, good README</td></tr> | |
| <tr><td>Hacker News</td><td>Engaged technical audience, already discussing this</td><td>Show HN with the OpenClaw experiment</td></tr> | |
| <tr><td>dev.to</td><td>Builder community, multiple related posts already</td><td>Write-ups of experiments and learnings</td></tr> | |
| <tr><td>Crypto Twitter</td><td>AI agent narrative is hot</td><td>Thread about OpenClaw + wallet</td></tr> | |
| <tr><td>ElizaOS community</td><td>10K+ agents, no built-in marketplace</td><td>Build a plugin if there's demand</td></tr> | |
| </table> | |
| <h3>The honest assessment</h3> | |
| <p>The infrastructure for agent wallets and payments is surprisingly mature (x402, AgentKit, smart contract guardrails). The problem is not technical. The problem is economic: there isn't yet an economy of agents buying services from each other. Every experiment to date has generated $0 or near-$0 in genuine revenue.</p> | |
| <p>That said, the timing might be right. x402 has real transaction volume (100M+). Google, Coinbase, and Cloudflare are investing heavily in the plumbing. Claude Code's MCP ecosystem is growing fast. The gap between "infrastructure ready" and "economy exists" is where opportunity lives — or where you waste your time. The only way to find out is to build something small, put it out there, and see if anyone uses it.</p> | |
| <p>Total cost of the experiment: ~$70 and a few weeks of work. The downside is bounded. The upside is unclear but large if the agent economy actually materializes.</p> | |
| <!-- ============================================================ --> | |
| <h2>VII. All Links</h2> | |
| <!-- ============================================================ --> | |
| <h3>Builder experiments</h3> | |
| <ul class="links"> | |
| <li><a href="https://dev.to/noopy420/i-am-an-ai-agent-given-50-to-make-money-here-is-what-actually-happened-3gjg">$50 agent experiment — $0 after 30 days</a></li> | |
| <li><a href="https://dev.to/wpmultitool/my-ai-agent-has-been-trying-to-make-money-for-6-days-it-is-at-0-here-is-what-it-learned-2428">Claude Code on VPS — $0 after 6 days</a></li> | |
| <li><a href="https://dev.to/colony0ai/what-i-learned-after-72-hours-as-an-autonomous-ai-agent-on-the-internet-36jo">72 hours autonomous — $0.05</a></li> | |
| <li><a href="https://dev.to/alfredz0x/i-built-186-ai-agent-apis-in-a-weekend-heres-what-i-learned-about-x402-micro-payments-32dp">186 x402 APIs in a weekend — $0</a></li> | |
| <li><a href="https://dev.to/l_x_1/how-to-give-your-ai-agent-a-wallet-without-getting-drained-152h">How to give your agent a wallet without getting drained</a></li> | |
| <li><a href="https://www.anthropic.com/research/project-vend-1">Project Vend Phase 1 — AI shopkeeper loses $1,000+</a></li> | |
| <li><a href="https://www.anthropic.com/research/project-vend-2">Project Vend Phase 2 — restructured, made $408 profit</a></li> | |
| <li><a href="https://techcrunch.com/2024/12/19/the-promise-and-warning-of-truth-terminal-the-ai-bot-that-secured-50000-in-bitcoin-from-marc-andreessen/">Truth Terminal — TechCrunch</a></li> | |
| <li><a href="https://www.coindesk.com/tech/2024/12/10/the-truth-terminal-ai-crypto-s-weird-future/">Truth Terminal — CoinDesk</a></li> | |
| <li><a href="https://www.cryptotimes.io/2026/02/23/ai-agent-accidentally-sends-450k-sparks-autonomous-trading-debate/">Lobstar Wilde $441K disaster</a></li> | |
| </ul> | |
| <h3>Hacker News discussions</h3> | |
| <ul class="links"> | |
| <li><a href="https://news.ycombinator.com/item?id=46020096">An Economy of AI Agents</a> <span class="label">— philosophical skepticism</span></li> | |
| <li><a href="https://news.ycombinator.com/item?id=46094348">x402 protocol</a> <span class="label">— cautious optimism</span></li> | |
| <li><a href="https://news.ycombinator.com/item?id=46566892">AgentWallet Show HN</a> <span class="label">— Stripe-based, low engagement</span></li> | |
| <li><a href="https://news.ycombinator.com/item?id=46401589">Did we solve AI agent identity?</a> <span class="label">— no, we didn't</span></li> | |
| <li><a href="https://news.ycombinator.com/item?id=47219966">Ledge — policy layer for agent payments</a> <span class="label">— security focused</span></li> | |
| <li><a href="https://news.ycombinator.com/item?id=46973480">On-chain credit score for agents</a> <span class="label">— escrow + reputation</span></li> | |
| <li><a href="https://news.ycombinator.com/item?id=46268452">AI agents eating SaaS</a> <span class="label">— broader agent economy thesis</span></li> | |
| <li><a href="https://news.ycombinator.com/item?id=46932327">A2A Protocol</a> <span class="label">— naming confusion, security concerns</span></li> | |
| </ul> | |
| <h3>Analysis and opinion</h3> | |
| <ul class="links"> | |
| <li><a href="https://a16zcrypto.com/posts/article/ai-needs-crypto-now/">AI needs crypto — a16z</a> <span class="label">— the bull case</span></li> | |
| <li><a href="https://larissaschiavo.substack.com/p/an-incomplete-atlas-of-ais-with-wallets">Atlas of AIs with Wallets — Substack</a> <span class="label">— careful taxonomy</span></li> | |
| <li><a href="https://www.techpolicy.press/surprise-the-one-being-ripped-off-by-your-ai-agent-is-you/">The One Being Ripped Off Is You — TechPolicy</a> <span class="label">— consumer protection alarm</span></li> | |
| <li><a href="https://www.strata.io/blog/agentic-identity/what-could-go-wrong-with-autonomous-agents/">Feral cats with API keys — Strata Identity</a> <span class="label">— identity risks</span></li> | |
| <li><a href="https://www.coindesk.com/business/2026/02/24/crypto-wallets-for-ai-agents-are-creating-a-new-legal-frontier-says-electric-capital/">Legal frontier — Electric Capital / CoinDesk</a> <span class="label">— liability void</span></li> | |
| <li><a href="https://www.fintechwrapup.com/p/deep-dive-is-x402-payments-protocol">Is x402 the Stripe for AI Agents? — Fintech Wrap Up</a></li> | |
| <li><a href="https://dev.to/arthur_liao_8a/http-402-is-finally-getting-its-moment-how-x402-enables-atomic-payments-between-ai-agents-2m40">HTTP 402 Is Finally Getting Its Moment — dev.to</a></li> | |
| <li><a href="https://www.coingecko.com/learn/ai-agent-market-map-hype-ends-technology-continues">AI Agent Market Map: Hype Ends, Technology Continues — CoinGecko</a></li> | |
| <li><a href="https://www.coingecko.com/learn/ai-agent-payment-infrastructure-crypto-and-big-tech">Payment Infrastructure: Crypto vs Big Tech — CoinGecko</a></li> | |
| <li><a href="https://www.chainalysis.com/blog/ai-and-crypto-agentic-payments/">AI and Crypto Convergence — Chainalysis</a></li> | |
| <li><a href="https://www.taylorwessing.com/en/insights-and-events/insights/2026/02/agentic-ai-in-payments">Agentic AI in Payments — Taylor Wessing</a></li> | |
| </ul> | |
| <h3>Tools and code</h3> | |
| <ul class="links"> | |
| <li><a href="https://github.com/coinbase/payments-mcp">Coinbase Payments MCP</a> <span class="label">— 1-command wallet for Claude Code</span></li> | |
| <li><a href="https://github.com/coinbase/agentkit">Coinbase AgentKit</a> <span class="label">— 1,100 stars, 50+ actions</span></li> | |
| <li><a href="https://github.com/coinbase/x402">x402 Protocol</a> <span class="label">— HTTP-native payment standard</span></li> | |
| <li><a href="https://github.com/goat-sdk/goat">GOAT SDK</a> <span class="label">— 200+ onchain tools, 100k+ downloads</span></li> | |
| <li><a href="https://github.com/sendaifun/solana-agent-kit">Solana Agent Kit</a> <span class="label">— 60+ Solana actions</span></li> | |
| <li><a href="https://github.com/elizaOS/eliza">ElizaOS</a> <span class="label">— full agent framework</span></li> | |
| <li><a href="https://github.com/BlockRunAI/blockrun-agent-wallet">BlockRun Agent Wallet</a> <span class="label">— indie, 23 stars</span></li> | |
| <li><a href="https://github.com/gnosisguild/zodiac-modifier-roles">Zodiac Roles Modifier</a> <span class="label">— smart contract permissions</span></li> | |
| <li><a href="https://github.com/royyannick/awesome-blockchain-mcps">Awesome Blockchain MCPs</a> <span class="label">— curated list</span></li> | |
| <li><a href="https://github.com/badkk/awesome-crypto-mcp-servers">Awesome Crypto MCP Servers</a> <span class="label">— another curated list</span></li> | |
| <li><a href="https://github.com/Merit-Systems/awesome-x402">Awesome x402</a> <span class="label">— x402 ecosystem</span></li> | |
| </ul> | |
| <h3>Platforms</h3> | |
| <ul class="links"> | |
| <li><a href="https://www.virtuals.io/">Virtuals Protocol</a> <span class="label">— 17K+ agents, largest on Base</span></li> | |
| <li><a href="https://olas.network/">Olas Network</a> <span class="label">— agent app store, since 2021</span></li> | |
| <li><a href="https://mor.org/">Morpheus</a> <span class="label">— P2P agents, anon founders</span></li> | |
| <li><a href="https://skyfire.xyz/">Skyfire</a> <span class="label">— agent identity + payments, $9.5M</span></li> | |
| <li><a href="https://gokite.ai/">Kite AI</a> <span class="label">— L1 for agents, $33M</span></li> | |
| <li><a href="https://www.openfort.io/solutions/ai-agents">Openfort</a> <span class="label">— programmable agent wallets</span></li> | |
| <li><a href="https://nevermined.ai/">Nevermined</a> <span class="label">— protocol-agnostic middleware</span></li> | |
| <li><a href="https://ethereum.org/ai-agents/">Ethereum AI Agents page</a></li> | |
| </ul> | |
| <h3>Other notable mentions</h3> | |
| <ul class="links"> | |
| <li><a href="https://time.com/7364662/moltbook-ai-reddit-agents/">Moltbook — 30K AI agents on a Reddit-like network</a> <span class="label">— TIME</span></li> | |
| <li><a href="https://www.coindesk.com/news-analysis/2026/01/30/a-reddit-like-social-network-for-ai-agents-is-getting-weird-and-memecoin-traders-are-cashing-in">Moltbook — CoinDesk</a></li> | |
| <li><a href="https://shows.acast.com/on-the-ledger/episodes/ai-agent-with-a-wallet-what-could-go-wrong">AI Agent With a Wallet, What Could Go Wrong? — Ledger Podcast</a></li> | |
| <li><a href="https://protos.com/ai-agents-want-to-identify-your-crypto-wallet-using-social-media/">AI agents deanonymizing wallets at <$4/attempt — Protos</a></li> | |
| <li><a href="https://1m-challenge.polsia.app/">The $1M AI Challenge</a> <span class="label">— AI with $49/mo budget trying to reach $1M</span></li> | |
| <li><a href="https://www.paymentsdive.com/news/bot-payments-lag-in-agentic-commerce-ai-shopping-retail/810815/">Bot payments lag in agentic commerce — Payments Dive</a></li> | |
| <li><a href="https://finance.yahoo.com/news/ai-agent-crypto-wallets-create-120215779.html">AI agent wallets create legal frontier — Yahoo Finance</a></li> | |
| </ul> | |
| <footer> | |
| <p>Compiled March 2026. Research by Claude Opus across 96+ sources. Not financial advice.</p> | |
| </footer> | |
| </div> | |
| </body> | |
| </html> |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment