AI agents are moving from demos to real commercial workflows, but the money flowing through those systems remains smaller than some of the market’s louder claims suggest. That is the central takeaway from recent commentary and research tied to Andreessen Horowitz, or a16z, which points to a fast-rising adoption curve for agentic tools even as payment volumes in the category remain early and uneven. For U.S. businesses, the message is clear: AI agents are gaining traction, but the infrastructure for trusted, scalable payments is still being built.
A reality check for AI agent payments
The phrase “AI agent payment volumes lower than reported, but adoption is growing: a16z” captures a broader shift now underway in enterprise technology. Investors, payment companies, and software vendors increasingly describe a future in which AI agents can search, negotiate, buy, reconcile, and even pay on behalf of users or businesses. Yet a16z’s own fintech commentary shows that the market is still in an early formation stage, with core questions around authorization, fraud, liability, and payment rails still unresolved.
In its May 2025 fintech newsletter, a16z said major payment networks and platforms were already moving into the space. The firm noted that, within a single month, Mastercard and Visa launched AI agent payment solutions, while PayPal introduced its first MCP server. That activity signals momentum at the infrastructure layer, but it does not necessarily mean large-scale transaction volume has already materialized.
That distinction matters. In emerging technology markets, headline excitement often runs ahead of measurable commercial throughput. AI agents may be technically capable of initiating transactions, but broad-based payment volume depends on merchant acceptance, identity verification, consumer trust, compliance controls, and clear rules on who is liable when something goes wrong. a16z’s framing suggests those foundations are still being assembled.
Why adoption is still accelerating
Even if payment volumes are lower than some public narratives imply, adoption of AI agents is clearly growing across adjacent categories. a16z’s enterprise AI research, published in early 2026, found that 78% of surveyed enterprise CIOs were using OpenAI models in production, while 81% of enterprises were using three or more model families in testing or production, up from 68% less than a year earlier. That indicates a broad willingness among companies to operationalize AI systems, including agent-like workflows, even before every monetization layer is mature.
The same report found that all three leading model providers tracked by a16z showed strong absolute spend growth, even as market share shifted among vendors. For U.S. companies, that means AI budgets are expanding in practical terms, not just in pilot programs. It also suggests that businesses are increasingly comfortable integrating AI into customer support, software development, knowledge management, and data analysis, all of which can evolve into more autonomous agent behavior over time.
Voice is another signal. In a16z’s 2025 update on AI voice agents, the firm said the market “exploded” in the second half of 2024 and noted that companies building with voice represented 22% of the most recent Y Combinator class cited in the report. a16z also said that since 2020 there had been 90 voice agent companies in YC, with acceleration continuing into newer cohorts. Those figures do not measure payments directly, but they do show that agent adoption is broadening across commercial use cases where transactions could eventually follow.
The gap between usage and transaction volume
One reason payment volumes may appear overstated is that “adoption” and “payments” are not the same thing. A company can deploy an AI agent to answer customer questions, qualify leads, summarize documents, or automate internal workflows without allowing that agent to move money. In many cases, businesses deliberately stop short of autonomous payments because the control environment is not mature enough.
a16z’s own writing highlights the barriers. The firm points to three major issues in agentic payments:
- Fraud and Know Your Agent: businesses need to verify that an agent is authorized and not malicious.
- Liability: chargebacks and transaction reversals raise questions about who bears the loss.
- Payment rails: agent-to-agent and machine-to-machine transactions may require different infrastructure, including wallets or stablecoin-based systems.
These are not minor technical details. They are the difference between a useful assistant and a trusted economic actor. Until those issues are standardized, payment volume is likely to remain concentrated in narrow pilots, controlled environments, or low-risk use cases rather than in mass-market autonomous commerce. That helps explain why adoption can rise quickly while actual payment throughput remains modest.
AI agent payment volumes lower than reported, but adoption is growing: a16z
The strongest reading of the a16z thesis is not that the market is weak. It is that the market is early. The firm’s fintech team has identified a growing stack of companies building around agent payments, including startups focused on billing, agent-specific payment rails, and wallets for AI systems. a16z specifically highlighted Paid.ai, Nekuda, and Payman as examples of companies trying to solve monetization and transaction challenges in the category.
At the same time, a16z’s broader enterprise and voice research shows that businesses are already spending heavily on AI systems that can become more agentic over time. Enterprises are using multiple model families, trust in frontier labs is rising, and total AI spend continues to grow. In that context, lower-than-expected payment volume does not undermine the long-term thesis. Instead, it suggests the commercial stack is developing in layers: first model adoption, then workflow automation, then higher-stakes actions such as payments.
There is also evidence that pricing and economics are changing fast enough to support wider deployment. a16z noted that OpenAI cut GPT-4o realtime API pricing in December 2024 by 60% for input and 87.5% for output. Lower inference costs can make agentic products more viable, especially in high-volume customer service and operations settings. But lower model costs alone do not create payment volume; they simply make it cheaper to run the software that may eventually initiate transactions.
What this means for U.S. businesses and investors
For U.S. merchants, banks, fintech firms, and enterprise software buyers, the current phase of the AI agent market calls for measured optimism. The opportunity is real, but the most credible near-term gains may come from workflow automation rather than fully autonomous purchasing. Companies that treat agent payments as an immediate mass-market revenue stream may be disappointed. Companies that build controls, identity systems, and narrow use cases first may be better positioned.
Several implications stand out:
- Payments infrastructure providers have a chance to define standards for agent authorization and liability.
- Enterprises can capture value now by deploying agents in support, operations, and coding before expanding into payments.
- Investors may need to separate adoption metrics from transaction metrics when valuing startups in the sector.
- Regulators and compliance teams are likely to play a larger role as agents move closer to financial decision-making.
The market may also split by use case. In lower-risk environments, such as internal procurement or tightly scoped B2B workflows, agent payments could scale faster. In consumer commerce, where fraud, disputes, and consent are more complex, adoption may take longer. That is an inference based on the issues a16z identifies around fraud and liability, rather than a direct forecast from the firm.
A market growing up, not slowing down
The current state of AI agents looks less like a collapse of expectations and more like a normalization of them. The loudest claims about payment volume may be ahead of the facts, but the underlying adoption trend remains strong. a16z’s research points to expanding enterprise AI usage, rising trust in production deployments, and growing startup formation around agentic products and infrastructure.
That combination is important. Technology markets often mature in stages, and payments tend to be one of the last layers to scale because they require trust, compliance, and clear accountability. If AI agents continue to gain ground in software development, customer support, voice interfaces, and business operations, then payment activity may follow as the rails improve.
Conclusion
AI agent payment volumes may be lower than some reports and market narratives suggest, but the broader adoption story remains intact. a16z’s recent work shows a market where enterprises are spending more on AI, deploying more models in production, and exploring new agentic workflows across industries. In that environment, payments are not the first proof point of success; they are the next frontier. For the U.S. market, the near-term winners are likely to be companies that build trust, controls, and practical use cases before chasing headline transaction numbers.
Frequently Asked Questions
What does a16z mean by AI agent payments?
It refers to systems that let AI agents initiate or complete transactions on behalf of users or businesses, including purchases, billing actions, or machine-to-machine payments. a16z’s fintech writing also discusses wallets, payment rails, and authorization layers built specifically for agents.
Why are AI agent payment volumes lower than reported?
A likely reason is that many companies have adopted AI agents for workflow automation without giving them permission to move money. a16z highlights unresolved issues around fraud, identity, and liability that can limit real transaction volume.
Is adoption of AI agents actually growing?
Yes. a16z’s enterprise AI research says 78% of surveyed enterprise CIOs use OpenAI models in production, and 81% of enterprises use three or more model families in testing or production, up from 68% less than a year earlier.
Which sectors are adopting AI agents fastest?
Based on a16z’s research, strong activity is visible in enterprise software, customer support, software development, data analysis, and voice-based applications. The firm also points to fintech, healthcare, and operations as active areas for voice agent startups.
Are major payment companies already entering the market?
Yes. a16z said that Mastercard and Visa launched AI agent payment solutions and that PayPal introduced its first MCP server within a recent one-month period covered by its 2025 fintech newsletter.
What should U.S. businesses watch next?
They should watch for clearer standards on agent identity, fraud prevention, liability, and merchant acceptance. Those factors are likely to determine when AI agent adoption translates into meaningful payment volume.
