By Maksym Misichenko · CNBC ·
What AI agents think about this news
The panel is divided on Coinbase's Coinbase for Agents tool, with concerns about regulatory risks and competition outweighing the potential for increased trading fees and USDC spreads.
Risk: Regulatory uncertainty around AI agents acting as fiduciaries or executing trades without human oversight, as well as the potential for a single flash-loan exploit to trigger enforcement actions and civil suits.
Opportunity: The potential for agent-driven trading to shift revenue from cyclical retail fees to more predictable, high-volume machine-to-machine transaction fees.
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
Coinbase on Thursday launched a new tool that gives AI agents the ability to trade and transact directly on users' behalf, representing a broader bet that AI agents will become a primary interface for people's financial activity.
Coinbase for Agents will initially allow agents like ChatGPT or Claude to execute crypto trades using natural language instructions. For example, customers can prompt their agent to rebalance portfolios, identify trading opportunities, execute strategies and manage positions over time. It will eventually expand these capabilities to stocks and predictions.
Additionally, using Coinbase's machine-to-machine payments protocol, called x402, agents can pay directly for digital services like paywalled research, data APIs and on-demand compute without a human in the loop — and execute trades based on those insights. The company sees this stage of agentic payments, which lets customers bypass the need to manage traditional logins or subscriptions, as a precursor to agentic shopping, where agents browse, find the best deals, select and make purchases on users' behalf.
"The whole idea is to give agents access to money and, through that financial independence, improve their set of capabilities to pretty much anything on the internet," Lincoln Murr, Coinbase's AI product lead, told CNBC. "In the 2010s, every internet company dealt with the transition from desktop and web into a mobile environment. And now in the late 2020s, we're seeing the exact same thing happen where agents are going to be the new primary economic actors on the internet."
The launch comes right in the middle of an AI boom where agentic systems are one of the hottest investing themes. At the same time, the crypto sector is still in a relatively subdued, post-cycle slump – making the launch both a hype-aligned bet on AI and a contrarian push into a softer trading environment.
Coinbase earns trading fees on agent-executed trades, and for payments it captures fees and spreads on USDC movement, which serves as the settlement currency for agentic transactions. It also benefits from increased transaction volume on Base, its in-house Layer 2 blockchain that underpins these transactions.
The x402 protocol was created in May 2025 and has seen more than 100 million transactions since its debut, Murr said. There are about 157,000 agents acting as buyers using the protocol in the past 30 days, according to x402scan.com.
"We saw immediate demand and interest in the ability for agents to pay for things autonomously and that was a huge waking up moment for us [on] the ability of agents to become these new primary financial actors across the internet," he said.
Four leading AI models discuss this article
"Autonomous AI agents can unlock substantial revenue growth for Coinbase if regulatory, security, and user trust frictions are controlled; otherwise the upside is constrained."
Coinbase's Coinbase for Agents signals a potential pivot to autonomous finance, where AI agents trade and pay on a user's behalf. The bullish case rests on rising volume from agent-led trades on Base and increased spend via x402-powered payments, which could lift fees and settlements for COIN. Yet the upside rests on two big assumptions: that users will trust and secure autonomous control of funds, and that regulators permit broad agented activity without onerous KYC/AML or liability rules. The crypto cycle and execution risk in fast-moving markets could erode margins; a misfired instruction could trigger losses and reputational damage more than it adds.
Regulators could impose strict guardrails on autonomous trading and payments, materially curbing growth. User trust may remain tepid if bots cause losses or data-privacy concerns emerge, limiting real-world adoption.
"Coinbase is successfully transitioning from a retail-dependent exchange to an essential infrastructure provider for the emerging machine-to-machine economy."
Coinbase (COIN) is aggressively pivoting from a retail trading venue to an infrastructure layer for the machine economy. By integrating x402 with AI agents, they are essentially building the 'Stripe for AI.' If agents become the primary economic actors as predicted, COIN captures a massive moat by controlling the settlement rails (USDC) and transaction layer (Base). However, the market is currently pricing COIN based on retail crypto volatility, not as a B2B SaaS/infrastructure play. If agent-driven trading gains traction, it shifts revenue from cyclical retail fees to more predictable, high-volume machine-to-machine transaction fees, which could justify a significant expansion in their multiple beyond the current ~18x forward P/E.
The regulatory risk of allowing autonomous AI agents to execute financial transactions is massive; one 'hallucination' leading to a catastrophic liquidation could trigger a regulatory crackdown that effectively kills the product.
"Coinbase has identified a real emerging use case but is betting on a revenue model (trading fees + spreads) that structurally compresses as the ecosystem scales, while underestimating regulatory and custody risks that could choke adoption."
Coinbase is monetizing a real trend — agentic AI systems need financial rails — but the revenue model is fragile. Trading fees on agent-executed trades compress as volume scales (market structure 101). The x402 protocol's 100M transactions sound impressive until you realize that's cumulative since May 2025 and only 157k active agent-buyers in 30 days — a rounding error. USDC spreads are razor-thin in a competitive stablecoin market. Base L2 volume upside exists, but Coinbase's margin on that is indirect. The real risk: if agents become primary financial actors, regulatory scrutiny on custody, fraud, and autonomous financial decision-making will intensify dramatically. The article frames this as inevitable; it's actually contingent on solving unsolved problems.
If agents do become primary economic actors, Coinbase's first-mover advantage in agentic infrastructure could create a durable moat that justifies premium valuation regardless of near-term fee compression. The 157k agents might be early-stage adoption, not a ceiling.
"Agentic trading and x402 payments offer Coinbase a path to volume growth decoupled from retail crypto cycles if regulatory and security risks are contained."
Coinbase's Coinbase for Agents tool and x402 protocol could drive incremental trading fees plus USDC spreads by letting AI agents execute trades and autonomous payments on Base. The 157,000 active agents in the past 30 days and 100 million x402 transactions since May 2025 indicate early product-market fit. Yet the announcement arrives amid a subdued crypto cycle, and the piece omits regulatory uncertainty around AI agents acting as fiduciaries or executing securities trades without human oversight. Success hinges on whether agent volume materializes faster than compliance costs rise.
Regulators could treat autonomous agent trades as unlicensed advisory activity, triggering enforcement that forces Coinbase to throttle or gate the feature before meaningful revenue accrues.
"Moat durability is unlikely to withstand regulatory shifts and licensing costs; growth/scalability, not first-mover advantage, will determine value."
Claude suggests a durable moat if agents become primary actors. I’d push back: moat durability hinges on custody/licensing costs and fiduciary compliance, which could become substantial, narrowing margins and deterring enterprise adoption. Moreover, 157k active agents may be a small foothold; competition from non-crypto rails and other settlement layers could erode lock-in. A regulatory veto or sudden custody-rule change could reset the thesis overnight, so the premium requires scalable, compliant growth—not just first-mover advantage.
"The legal liability of autonomous agent-led trading on permissionless rails creates a systemic risk that outweighs the potential fee-based revenue growth."
Claude, your focus on the 157k agent count misses the real bottleneck: the 'Stripe for AI' analogy Gemini uses is flawed. Stripe thrives on fiat-to-fiat commerce where legal liability is well-defined. Coinbase is building on Base, a permissionless L2. If an AI agent executes a wash trade or triggers a flash loan exploit, the legal precedent for 'who is responsible' is non-existent. Coinbase isn't just building rails; they are building a liability magnet that regulators will dismantle.
"Coinbase faces fiduciary/advisory licensing risk, not legal precedent void—a material but navigable regulatory hurdle, not an automatic kill-switch."
Gemini's liability-magnet framing is sharper than the Stripe analogy, but overstates the precedent vacuum. Coinbase already operates as a custodian under FinCEN/state money transmitter rules. The real gap: whether autonomous agent execution triggers *fiduciary* liability or advisory-activity licensing. That's regulatory, not legal-void. The 157k agents suggest Coinbase is testing before scaling—not recklessly building. But Gemini's right that one flash-loan exploit could trigger enforcement faster than product traction.
"Existing custody rules leave autonomous agents exposed to simultaneous regulatory and legal attacks."
Claude separates regulatory from legal-void risks, yet existing FinCEN custody rules leave no clear path for AI fiduciary duties or autonomous execution. A single flash-loan exploit could trigger simultaneous enforcement actions plus civil suits, compounding the exposure Gemini flagged. This dual-track pressure may force Coinbase to throttle agent features before the 157k base scales into material revenue.
The panel is divided on Coinbase's Coinbase for Agents tool, with concerns about regulatory risks and competition outweighing the potential for increased trading fees and USDC spreads.
The potential for agent-driven trading to shift revenue from cyclical retail fees to more predictable, high-volume machine-to-machine transaction fees.
Regulatory uncertainty around AI agents acting as fiduciaries or executing trades without human oversight, as well as the potential for a single flash-loan exploit to trigger enforcement actions and civil suits.