The era of AI spending money for you—who is collecting taxes?

Author: Prathik Desai; Translation: BitpushNews

In March of this year, OpenAI shut down a feature that allowed AI agents to shop on behalf of users. In the five months since its launch, fewer than 30 Shopify merchants have actually used it. The core issue behind this was not the payment infrastructure but the lack of rules—at that time, there was no mechanism to ensure a seamless shopping experience. What exactly can an agent buy? Who should deduct sales tax? How to intercept fraud? Who handles returns and exchanges? These fundamental questions remained unresolved.

Configuring a wallet for an agent or building a payment infrastructure has long been technically straightforward. However, enabling individuals or businesses to authorize agents to spend their money in a secure, trustworthy, and governed manner is extremely challenging. Only “programmability” and “clear rules” can build a trustworthy ecosystem. The blank space in the governance layer (Governance Layer) is breeding enormous business opportunities within the agent economy.

Last year, AI agents completed 176 million transactions worldwide, with a total settlement amount of $73 million. Although this number seems insignificant now, McKinsey predicts that by 2030, the transaction volume mediated and matched by AI agents in global consumer commerce will soar to $3 trillion to $5 trillion.

Therefore, companies building this new economy are racing to seize the “governance layer”—including controlling spending limits, identity verification, and compliance enforcement—thus having full authority over which agents can be trusted with what amount of funds.

Today, we will deeply analyze: who is building the banking layer for these “bots,” and who, if they win this dominance battle, will reap astonishing dividends?

Why pursue full-chain vertical expansion?

The underlying economics of processing agent payments are extremely brutal. Over the past 12 months, the average transaction amount for AI agents was only 31 cents.

Imagine a micro-payment of 31 cents, which after passing through multiple layers of behind-the-scenes transaction chains, how much profit can the leading settlement institution still make? Using Stripe’s standard fee model (2.9% + 30 cents fixed fee), the profit left for merchants would be less than a tenth of a cent. Visa’s interchange fees would further eat up about one-third of that. In contrast, using blockchain Layer-2 (L2) stablecoin channels, the cost to process the same transaction is only $0.0001.

This extreme economic characteristic has laid a solid foundation for the application of cryptocurrencies at the settlement layer.

Currently, the settlement layer’s payment infrastructure is largely complete. Coinbase’s x402 protocol carries most of the 176 million transactions last year, with about 3,900 merchants accepting agent payments. Meanwhile, Stripe and Tempo have jointly developed a set of competing underlying standards—the Machine Payments Protocol (MPP)—which officially launched in March this year and has integrated with over 100 service systems. Additionally, Google, Visa, and Mastercard have also launched their own agent payment products during the same period. In just 12 months, five competing payment architectures have emerged in the market.

However, the core pain point of agent payments is: nobody can make money just by handling a 31-cent channel fee. The real commercial value is highly concentrated in two core areas: the interest generated from capital float (i.e., position interest), and the enforcement and control of agent payment rules (governance layer).

In last week’s article, we explained how companies can lock in stablecoin balances of AI agents through “wallet layer” control, earning interest on the deposited funds. But this is just one of many value layers to be captured. Another equally lucrative area is the “rule-making authority”—controlling how these deposited funds are spent.

These rules include: spending limits, agent identity verification, compliance enforcement, audit tracking, and responsibility for failed transactions. Currently, this governance layer is in a vacuum of competing factions.

In April this year, American Express launched the “Agent Purchase Protection” program, an insurance product specifically for reimbursement of erroneous purchases made by AI agents. This move implicitly admits that the current governance layer for AI agents is severely lacking in security. In this huge industry, expected to grow to $3–$5 trillion within five years, whoever can fill this governance gap will capture enormous value.

This is why major players are desperately fighting for the governance track.

But what dimension should this governance layer be built on? It could be a bank, an API for developers, or even a wallet.

Wallets: the natural governance gateway

Every dollar spent by an agent must flow through a wallet. This makes wallets an ideal choke point for enforcing spending limits, identity checks, and manual approvals. Once you control the wallet, you hold the key to governance. Third-party payment giant Stripe has long been keenly aware of this.

In June 2025, Stripe acquired Privy, a cutting-edge company specializing in embedding wallets for consumer-grade crypto applications. Through this acquisition, Stripe gained access to over 75 million wallets across more than 1,000 developer teams. Today, these wallets sit at the most critical “checkpoint”—all policies, limits, and manual authorizations must pass through here before any funds are transferred.

Additionally, Stripe quietly assembled a complete stack of agent payment technology. It acquired Bridge to handle stablecoin dispatch and fiat conversions; partnered with Paradigm to incubate Tempo, a Layer 1 blockchain focused on payments. As mentioned earlier, Stripe and Tempo jointly established the Machine Payments Protocol (MPP), setting open standards for how agents apply for, authorize, and settle funds.

Thanks to this financial tech stack tailored for agents, software systems can easily enable AI agents to check balances, pay bills, deposit funds, open virtual cards, and perform cross-border remittances. Agents can autonomously execute routine payments, and if any operation exceeds policy boundaries, the system automatically intercepts and escalates for manual review. Currently, the treasury account balance behind this system is supported by Privy’s non-custodial wallets distributed across over 150 markets worldwide.

Even Amazon, when it needs to empower its platform developers to “let AI agents spend money,” ultimately chooses wallet providers—Privy and Coinbase. It does not opt for traditional banking giants or card networks, but a wallet service provider founded just five years ago.

The underlying logic is: wallets are the ideal containers for implementing control points, allowing for appropriate manual intervention to ensure checks and balances.

Keyrock, a digital asset market maker, states in its industry report “Who Pays the Agent” that: “The future of agent commerce will settle into a compromise—agents will enjoy high autonomy but must operate within ‘hard boundaries’ enforced by cryptographic technology, with humans able to audit or revoke permissions at any time.”

This precisely describes Privy’s strategic position within Stripe’s ecosystem. Wallets are used to enforce behavioral boundaries for agents.

Governance strategies for agent wallets

In practical governance implementation, Privy offers two operational modes for agent wallets:

Full Autonomy Mode: The agent has absolute control over the wallet and executes transactions under compliance policies without human intervention. This mode is best suited for high-frequency trading bots and automated asset management agents.

Restricted Co-Management Mode: The ultimate ownership remains with the human user, but the user grants limited permissions to the agent, allowing it to act as a “co-signer” on the wallet. The user retains full authority to revoke the agent’s access at any time.

Stripe’s MPP protocol also adopts similar governance principles.

For high-frequency agent tasks, MPP innovatively introduces “Sessions.” Under this mechanism, an agent can be pre-authorized with a total spending budget, within which it can continuously push payments without repeated manual approval for each transaction. Currently, MPP supports sub-cent billing for large language model (LLM) inference and single-query precise billing for data APIs.

This highly granular governance is fundamentally unsupported by traditional card networks like Visa and Mastercard.

Vertical integration of the business landscape

Although Coinbase, with its x402 protocol, currently leads in total AI agent payment volume, Privy has a trump card unrelated to cryptography itself—its vast merchant distribution network derived from Stripe.

Coinbase has about 3,900 merchants accepting agent payments. However, each merchant on Coinbase corresponds to nearly a thousand merchants on Stripe. In February this year, Privy announced that as long as Stripe’s merchants agree to adopt machine payments, the agent economy can rapidly scale through Privy’s existing wallet network overnight. Stripe’s merchants don’t need to build complex crypto infrastructure from scratch.

As competition between Stripe and Coinbase intensifies, traditional financial giants are also rushing to expand their vertical tech stacks through aggressive acquisitions.

According to Keyrock’s comprehensive map, the entire agent payment ecosystem is divided into six core layers: settlement, wallet, routing, protocol, governance, and application. In this ecosystem, 179 industry projects are fiercely competing.

In terms of coverage, Coinbase and Stripe are the most dominant, each spanning five of the six layers; stablecoin giant Circle follows with four layers. In contrast, despite Google’s massive scale, it currently only reaches two layers, while payment leader Visa covers just one.

Over the past 12 months, these giants have invested over $8 billion to acquire these technological gaps:

Capital One paid $5.15 billion to acquire the AI-native software platform Brex.

Mastercard spent $1.8 billion to acquire crypto payment infrastructure BVNK.

These capital moves clearly signal that the industry has identified where the truly scarce assets are. Pure settlement infrastructure is becoming cheaper and more commoditized; the real profit lies in managing permissions, controlling budgets, and clarifying legal responsibilities—i.e., the governance layer.

This vertical integration also creates powerful compounding effects.

Whoever can firmly control the wallet choke point can logically set spending rules, intercept interest before funds move, monopolize “trust access” for certain merchants and applications, and extract substantial ecosystem service fees. The distribution effects between Privy and Stripe exemplify this.

Looking at Coinbase’s ecosystem, the logic is similar: each agent payment via x402 on its own Layer 2 network, Base, generates continuous demand for USDC stablecoins, directly earning float revenue. These profits are then reinvested into its developer toolkit, AgentKit, which comes with session limits, single-transaction locks, and whitelist-only fund transfers. The more agents developed with AgentKit, the more x402 settlement transactions are generated, creating a virtuous cycle.

The undercurrents of investment among giants run even deeper.

Coinbase Ventures recently invested heavily in three leading independent governance startups: Catena Labs, Skyfire, and Payman. Circle’s co-founder Sean Neville founded Catena, and Circle has also made strategic investments in Skyfire. Top Silicon Valley VC firm a16z led early funding rounds for these companies. Even traditional giant Visa has supported Payman and formed strategic partnerships with Skyfire.

It’s clear that the same old players who built the global payment and settlement infrastructure are now turning to fund the agent governance layer. Their strategy is simple: if governance remains just an “embedded feature” within existing infrastructure (like Privy’s role in Stripe), they can maximize profits within the current ecosystem; if governance evolves into a new, independent industry segment, they can profit passively through their large investment portfolios.

What does controlling the governance layer mean for the size of the cake?

History repeatedly proves that merely providing channels or handling payments will never be the most profitable segment of the supply chain. As technology advances, financial infrastructure and pipelines tend to become commoditized. Once commoditized, industry profits shift to control points—those that decide whether a transaction is “permitted” and under what preconditions.

Many traditional industries have gone through this “commoditization” pain point.

Recall what happened when the internet shattered the profit barriers of traditional cable TV networks. All ISPs (broadband providers) overnight became interchangeable commodities. To avoid becoming mere pipes, telecom giants were forced into aggressive vertical expansion.

In India, top telecom players Jio and Airtel packed hundreds of TV news channels, half a dozen OTT platforms, unlimited free calls, set-top boxes, and free routers into basic broadband packages. In the US, AT&T spent $85 billion acquiring media giant Time Warner, transforming into a super-giant combining media content and telecom infrastructure. Their strategy was clear: tightly bundle top content like HBO, Warner Bros., and CNN with their vast distribution network to fend off streaming giants like Netflix and Amazon.

When the lowest layer—network access—becomes a cheap commodity, value shifts toward content, user relationships, and ecosystems that can hook consumers.

Similarly, in early crypto, a similar pattern played out.

Theoretically, settlement should happen on the underlying protocol chain (Ethereum, for example, as a shared ledger). But when Coinbase launched its faster, cheaper Layer 2 network Base, it began earning “sequencer revenue” from transactions on its own chain. Today, Coinbase earns nearly $60 million annually just from processing transactions on Base.

Those building agent payment channels are clearly studying these lessons thoroughly.

In previous articles, we explained that simply locking stablecoin balances between two transactions can generate substantial interest—creating a “free money” stream. This makes any company that first captures the “wallet layer” a robust cash cow.

Once you further dominate the “governance layer,” you unlock an even larger cash flow.

Looking at traditional finance: Visa processes $14.2 trillion annually, earning a take rate of about 0.28%. This fee includes not only hardware and processing costs but also a “trust premium”—the cost merchants and consumers pay for Visa’s fraud prevention, dispute resolution, and strict network rules.

If even a tiny fraction of this fee could be extracted in the agent economy, the wealth generated would be staggering. McKinsey’s conservative estimate for the 2030 agent economy at $3 trillion suggests that a mere 0.1% fee (a third of Visa’s total) would yield $3 billion annually.

To give a sense of scale: Coinbase’s total revenue in 2025 from subscriptions and services is about $2.8 billion. This means that just from fraud prevention, quota control, and governance fees in agent transactions, the revenue could surpass Coinbase’s entire current income from staking, custody, and membership services combined.

In conclusion, the players who will dominate in the future are those capable of vertically integrating across the entire tech stack—earning idle interest from wallet layer, transaction fees from settlement layer, and compliance and rule enforcement fees from governance layer.

In this wave of the agent financial voyage, full-stack vertical integration will be the ultimate business model for companies seeking long-term resilience and protection from commoditization.