Something quietly shifted in how the internet works. AI agents, software that can reason, plan, and act on your behalf, are starting to handle real money. Not in some distant future. Right now, in production, across some of the biggest platforms in tech. Google, Shopify, Stripe, OpenAI, Coinbase: they're all building infrastructure for a world where your AI assistant doesn't just find the right product but buys it for you. And on the other end of the spectrum, crypto wallets are giving agents the ability to hold, spend, and trade funds autonomously, with no bank account required. This is not a small shift. Edgar, Dunn & Company forecasts the agentic commerce market will grow from roughly $136 billion in 2025 to $1.7 trillion by 2030, a 66% annual growth rate. Morgan Stanley estimates agentic shoppers could represent $190 billion to $385 billion in U.S. e-commerce spending alone by the end of the decade. The money is moving. The question is whether we're ready for what comes with it.

Google and the Universal Commerce Protocol

In January 2026, Google announced the Universal Commerce Protocol (UCP) at the National Retail Federation conference. Co-developed with Shopify, Etsy, Wayfair, Target, and Walmart, UCP is an open standard designed to let AI agents operate across the full shopping journey: product discovery, comparison, checkout, and post-purchase support. Google's vision is ambitious. Its AI Mode in Search already lets users ask detailed, conversational questions about products. According to Google, shoppers using AI Mode ask questions that are 23 times longer than traditional search queries. That's not a tweak in behavior. That's a fundamentally different way of interacting with commerce. Google also introduced a Shopping agent built on Gemini that can process text, voice, and images, acting as a "digital concierge" that autonomously builds carts and executes purchases with user consent. Checkout is now embedded directly into Google Search and the Gemini app, powered by Google Pay and PayPal. The key word in all of this is embedded. The agent doesn't redirect you to a store. It completes the transaction where you already are.

Shopify and Stripe: making every AI conversation shoppable

Shopify has positioned itself as the backbone of agentic commerce for merchants. In its Winter '26 Edition, Shopify launched Agentic Storefronts, enabling merchants to sell products directly inside AI conversations on platforms like ChatGPT, Google Gemini, and Microsoft Copilot. The numbers back up the bet. Shopify reported 31% year-over-year revenue growth in 2025, and leadership credited much of that to AI-powered shopping. Purchases attributed to AI-powered search increased 11x across Shopify merchants. On the payments side, Stripe and OpenAI co-developed the Agentic Commerce Protocol (ACP), launched in September 2025. ACP powers Instant Checkout in ChatGPT, letting users buy from Etsy and Shopify merchants without ever leaving the chat. The protocol uses a Shared Payment Token that captures customer authorization while shielding financial data from the AI agent itself, a critical design choice for trust and security. Salesforce joined the effort in October 2025, integrating ACP into Agentforce Commerce so that thousands of merchants can plug into conversational AI checkout flows. What's emerging is a pattern: open protocols that let agents transact on behalf of users, with merchants keeping their existing payment infrastructure. The agent becomes a new sales channel, like a website or a mobile app, but one that lives inside a conversation.

Crypto: where agents don't need permission

The platforms above all operate within traditional financial rails. Agents act on behalf of a human, with explicit consent, through regulated payment processors. Crypto takes a fundamentally different approach. In the crypto world, an AI agent can own a wallet. It can hold tokens, pay for its own compute, trade assets, and even hire other AI agents, all without a bank account, a credit card, or a human intermediary. Coinbase launched Agentic Wallets, the first wallet infrastructure designed specifically for AI agents. The pitch is straightforward: agents need money that moves at the speed of code. Crypto rails enable instant, programmable transactions without the friction of traditional banking. Agents can spend, earn, and trade autonomously while operating within programmable guardrails set by their owners. Google Cloud also introduced the Agent Payments Protocol (AP2) in September 2025, specifically targeting on-chain payment flows for autonomous agents. The appeal is real. For certain use cases, like micropayments between AI services, cross-border transactions, or automated portfolio rebalancing, crypto offers speed and programmability that traditional finance simply can't match. But this freedom comes with serious risks.

The scam problem

Crypto's permissionless nature makes it a magnet for fraud, and AI has made the problem dramatically worse. Chainalysis estimates that crypto scams and fraud reached at least $14 billion on-chain in 2025, up from $12 billion the previous year. The firm projects the real number is closer to $17 billion once all illicit funds are identified. Impersonation scams alone posted 1,400% year-over-year growth. The most alarming finding: AI-enabled scams were 4.5 times more profitable than traditional scams. Groups that purchased AI tools averaged $3.2 million per swindle, compared to $719,000 for those not using AI. Deepfakes, synthetic identities, and automated phishing have industrialized what used to be labor-intensive con operations. TRM Labs reported an approximately 500% increase in AI-enabled scam activity over the past year, with illicit crypto volume totaling $158 billion in 2025, a 145% year-over-year increase. The irony is hard to miss. The same technology that makes agents useful for legitimate commerce, their ability to reason, adapt, and act autonomously, also makes them devastatingly effective tools for fraud. An AI agent with a crypto wallet doesn't check whether its instructions are ethical. It just executes. As one investor put it at a recent panel: "AI itself cannot be punished." When an agent with an independent wallet causes losses in transactions, lending, or commerce, there is still no clear answer on who bears responsibility.

Where this is heading

A few trends are becoming clear. Open protocols are winning. Google's UCP, Stripe and OpenAI's ACP, Shopify's MCP integrations: the biggest players are building open standards rather than walled gardens. This is partly pragmatic (agents need to work across platforms) and partly strategic (whoever defines the protocol shapes the market). Agents are becoming a sales channel. Just as businesses had to build websites in the 2000s and mobile apps in the 2010s, they'll need agent-ready storefronts in the late 2020s. The infrastructure is already being built. Gartner projects that by 2026, more than 20% of companies worldwide will use some level of autonomous AI in their operational processes. Trust infrastructure is the bottleneck. The hardest problem isn't technical, it's trust. How do you verify an agent's identity? How do you set spending limits? Who's liable when something goes wrong? Stripe's Shared Payment Token and Coinbase's programmable guardrails are early answers, but the regulatory frameworks haven't caught up. The gap between regulated and unregulated is widening. Google and Stripe are building agent commerce within existing financial regulations. Crypto is building it outside those regulations. Both approaches have legitimate use cases, but they carry very different risk profiles. The tension between permissioned and permissionless agent finance will define the next few years of this space. We're at the beginning of a fundamental rewiring of how money moves online. The agents are here. They have wallets. And increasingly, they're the ones deciding what to buy, where to buy it, and how to pay. The question isn't whether AI agents will handle money. They already do. The question is whether we'll build the right guardrails before the scale of the problem outpaces our ability to manage it.

References