Author Charlie, who has lived in Silicon Valley for 15 years, participating in and observing cutting-edge technology, focuses on how AI is reshaping financial services, payment networks, and capital markets. Fintechnize and the founder of Dune Road's Maucircle Lion, Venture Partner @ Generative Ventures. Formerly Head of Americas Payments at OSL, Vice President at cryptocurrency unicorn Strike (involved in El Salvador's Bitcoin legislation and responsible for the Bitcoin Lightning Network and stablecoin payment operations in Latin America), macro and currency analyst at trillion-dollar fund Franklin Templeton, and an early member of global payment giant Adyen. The opinions in this article are personal and not investment advice.

In the past week, several seemingly unrelated events occurred in the financial industry.

On August 25, Google Cloud launched Gemini Enterprise for Financial Services for financial institutions, further advancing AI into KYC, credit analysis, portfolio monitoring, bond issuance, and financial research, workflows that are truly close to core business.

Two days later, MoonPay announced the integration of PayBox into Kamino on Solana, allowing users to initiate lending and earn returns through interactive gateways like Claude or ChatGPT.

Meanwhile, the Dallas Fed published a seemingly unsexy study discussing what would happen if deposits could be moved in real time 24/7 and AI could automatically compare returns for businesses and individuals, challenging the deposit stickiness banks have always relied on.

The next day, the BIS mentioned Project Agorá, which had just completed a real funds test, while discussing stablecoins and tokenized deposits.

Seen individually, each event can be categorized into a familiar news classification: Google is corporate AI, MoonPay is crypto finance, Dallas Fed is bank research, and Project Agorá is the central bank and commercial banks exploring next-generation cross-border payment infrastructure.

But when viewed together, a more noteworthy thread emerges.

For the past two to three years, we have been discussing when AI will genuinely change finance. However, I increasingly feel that this question is not accurately posed. The real dividing line has never been when AI becomes “more knowledgeable about finance,” but rather when it starts being allowed to make financial decisions on behalf of people and can actually move money.

The difference between these two aspects might be greater than the leap from search engines to ChatGPT.

A few years ago, when I first began exploring the intersection of generative AI and Fintech, my overall judgment was relatively conservative. At that time, the most realistic applications in the finance industry were still customer service, reporting, fraud prevention, cost management, internal knowledge retrieval, and various tools to improve employee efficiency. These are certainly valuable, but they mainly stayed within the confines of “AI assisting next to the money” without truly interacting with the funds.

The reasons are quite simple. If a model writes a segment of market copy incorrectly, it can be revised; if a model mistakenly transfers $5 million for a business, clearly, just hitting “undo” won't resolve the issue. Financial institutions naturally have a much lower tolerance for black boxes compared to ordinary software companies. Accuracy, permissions, auditing, accountability, regulatory requirements—if any of these aspects are unresolved, no matter how intelligent the model is, it can only act as an assistant.

However, the notable change in recent months is not that the industry suddenly believes that “fully automated AI wealth management” will imminently arrive, but rather that more and more companies are beginning to break down the task of “empowering AI to manage money” into a series of smaller, more manageable authorizations.

When Robinhood launched Agentic Trading this year, it did not let a model take control of all user assets directly; instead, it specifically set up a separate account where AI could operate. By the end of the second quarter, there were nearly 100,000 such accounts, managing over $100 million in assets.

Ramp's approach is less flashy but may be more aligned with the evolution of real corporate finance. Its AI does not directly replace the CFO, but first reads the company's reimbursement, payment, and accounting policies, automatically processes high-certainty situations, and refers anomalies back to humans.

Recently launched by Rocket Money, Rowan reflects a similar trend: it no longer just provides users with a graphic of “how much you spent this month,” but can proactively discover subscriptions in the background, send reminders, help cancel services, negotiate bills, and even execute savings transfers according to preset rules.

When viewed collectively, these products reveal that the market is not jumping from “manual operation” to “fully autonomous machines,” but rather progressing along a more realistic path: first observing, then suggesting, then preparing actions, then human approval, and finally gradually delegating those clearly defined, risk-controlled actions to machines.

This is also why I believe that the truly important concept of Agentic Finance is not “automation” but “authorization.”

Today, when an employee uses a company credit card, it is already a form of financial authorization. Of course, they cannot spend all the money in the company's bank account, but they can freely spend within certain merchant categories, amounts, and approval rules. Fund managers, corporate financial officers, and purchasing managers are essentially managing funds on behalf of others within a predefined authorization framework.

The real role of the AI Agent is not to invent a brand new financial system but to translate this authorization relationship, which we have long taken for granted, into rules that machines can understand and execute.

For example, a company could instruct a financial AI agent: it must maintain at least 30 days of operating cash; remaining funds can only be placed in designated banks or high-rated products; no foreign exchange risk; any fund transfer over $500,000 must be manually approved; only when the yield difference exceeds 20 basis points is it worth switching; any newly emerging counterparties must undergo manual review.

Once these rules can be directly understood by machines, the nature of the situation changes fundamentally. The AI does not need to gain unlimited power or become an entirely unregulated “robot hedge fund.” It just needs to operate autonomously within defined boundaries.

This is also why the common direction among companies like Google, MoonPay, Visa, and Mastercard, as they engage in Agent payments, is becoming increasingly evident. On the surface, everyone is discussing how AI makes payments, but the real focus is on permissions: who authorizes, how much is authorized, until when, which merchants and accounts can be accessed, under what circumstances must processes be halted to seek human confirmation, and how to maintain a comprehensive audit trail once a problem arises.

Thus, I am increasingly convinced that one of the most important products in the future of Agentic Finance may not necessarily be a smarter large model but a sufficiently reliable “financial authorization system”: it can translate a very vague human instruction—like “help me manage the company's idle funds”—into hundreds of rules that machines can strictly execute.

Viewing MoonPay's recent developments over the past six months from this perspective is far more interesting than merely examining its collaboration with Kamino.

In March, MoonPay launched the Open Wallet Standard, focusing on how AI can safely use wallets and obtain limited signing permissions without touching private keys; in May, it acquired DFlow to add transaction execution infrastructure; in June, it acquired Entendre to enhance reconciliation, finance, fund management, and reporting capabilities; in July, it launched PayBox, integrating cards and wallets directly into interactive environments like Claude and ChatGPT; and by August, it connected lending and earnings management through Kamino.

Individually, each step may not seem groundbreaking. However, taken together, they resemble assembling a foundational capability aimed at AI Agents in finance: first addressing identity and permissions, then executing, followed by financial operations, and ultimately starting to touch on credit, lending, and asset allocation.

This is also the real distinction between Agentic Payments and Agentic Finance.

The former answers: How does AI pay for me?

The latter asks: How does AI manage my balance sheet for me?

Once we enter the second question, payment becomes just one of many actions. Where to hold cash, when to borrow money, what assets to collateralize, how much return idle funds generate, when to rebalance, how to manage liquidity and risk—these decisions that used to require continuous judgment from corporate finance teams, bankers, or investment managers may soon fall under machine execution.

Yet there remains an infrastructural contradiction here.

If AI can calculate the best solution every second, but money itself can only move at the speed of bank operating hours, cross-border correspondent banks, batch settlement, and manual reconciliation, then even the wisest Agent has limited significance. It’s like a high-performance sports car constrained to a bumpy country road.

This is why Project Agorá deserves to be included in this discussion.

Project Agorá is not an AI project per se. It is an experiment jointly promoted by the BIS Innovation Hub and the Institute of International Finance, researching how to embed tokenized commercial bank deposits and central bank reserves into a multi-currency, programmable platform to improve wholesale cross-border payments.

The real connection between it and Agentic Finance is not that “BIS has started doing AI,” but that it is investigating another piece of financial infrastructure that AI ultimately needs to access: how money itself can become easier for machines to read, combine, and execute.

In July, 28 financial institutions and central banks completed real fund tests in a controlled environment, involving Swiss Francs, Euros, Pounds, Yen, Korean Won, and US Dollars. The scale of the tests was quite small, far from being production-ready, but it at least validated one thing: commercial bank deposits and central bank currencies do not inherently have to exist only within today’s slow and segregated systems; they can just as easily enter an environment that supports conditionally triggered, atomic settlement, and automatic execution of rules.

The BIS itself remains quite cautious about this. Issues such as how systems will interconnect, how legal final settlements are recognized, accountability for smart contracts in case of issues, cybersecurity, governance, and how to transition decades-old systems all remain far from resolved. Therefore, to say that Agorá represents “the next generation of global financial systems has already been built” is clearly exaggerated.

However, it at least proves a point that can often be overshadowed by crypto narratives: programmable money does not necessarily equate to stablecoins and does not inherently mean bypassing banks. Traditional bank deposits themselves may also become financial assets callable by software.

The truly interesting aspect begins precisely here.

Because banks are currently pouring resources into eliminating friction in the flow of funds, but this friction has historically been part of the bank's business model to some extent.

This is the most noteworthy aspect of the Dallas Fed article.

Why are a bank's deposits valuable?

Of course, there are many lofty reasons. A company places $50 million in JPMorgan not just out of convenience, but because it may be tied to credit lines, foreign exchange, payroll, cash management, custody, and years of customer relationships. These factors will not just vanish because a rival bank's interest rate rises by 10 basis points.

However, another reason that is less grand but equally real: moving money is just too cumbersome.

Today, if a company has $2 million temporarily idle in an account, Bank A offers 4.20%, while Bank B offers 4.35%. Is it worth the finance team checking every morning for just 15 basis points? Is it worth logging in, comparing counterparty risk, making funds transfers, considering settlement times, and running another round of reconciliation? They must also ensure tomorrow's payroll and supplier payments won't be affected.

Most of the time, the answer is simply no.

The money just stays there.

This highlights an asset in the financial system that rarely gets discussed independently: human inertia.

The Dallas Fed's article states directly that the stickiness of operational deposits partly comes from the real friction of funds not being able to be rapidly reallocated. If real-time settlement, tokenized deposits, and AI Agents all mature in the future, clients seeking higher returns could theoretically switch funds with almost no action on their part.

AI wouldn't see a 15 basis points difference as “not worth the hassle.”

It doesn’t hold Monday morning meetings, doesn’t forget, nor does it delay tasks to the next week because it is busy today. For it, the marginal attention cost of continually comparing different financial products is almost zero.

This does not mean all corporate deposits will behave like hot money, switching banks every five minutes. Client relationships, credit lines, regulatory requirements, and risk management still exist, and switching funds will never actually reach zero cost.

But the point is, it doesn’t need to evolve to such extremes to impact banks.

Dallas Fed conducted a sensitivity analysis based on the US banking system's nearly $17 trillion in related deposits. If the average tenure of deposits shortens by 10%, the banking system's capacity to handle maturity transformation could potentially decrease by about $580 billion in “10-year duration risk exposure”; if deposit rates become 10% more sensitive to changes in market rates, the duration risk banks can take on may reduce by about $700 billion.

This does not mean AI will suddenly reduce the US banking system's lending capacity by $700 billion. Simplifying this number to say “lending capacity reduced by $700 billion” is inaccurate.

What it truly indicates is another matter: deposit stickiness itself has enormous economic value.

If in the future, programmable money and AI Agents merely weaken this stickiness slightly, the cost and duration structure of bank liabilities could change accordingly.

In the past, when discussing programmable money, the benefits were often highlighted: faster settlements, fewer reconciliations, lower costs, 24/7 operation, and the ability to directly embed payment conditions into transaction logic.

But there is another side that has rarely been emphasized: once money becomes easier to move, those holding it gain more options.

For asset holders, this is efficiency; for institutions relying on this money as a stable funding source, it signifies more intense competition.

In order for money to operate like software, banks may have to accept one outcome: money begins comparing prices at software speed.

This brings the issue to another direction I have been contemplating recently.

In the past, I always believed finance is essentially a distribution business.

Banks compete for who gets salaries deposited directly, credit card companies vie for who becomes consumers' most-used card, brokerage firms fight for who occupies investors’ mobile screens, and wealth management institutions contend for client relationships. Many fintech products spend billions to acquire customers, fundamentally competing for the same thing: who can position themselves between users and financial products.

Because whoever controls the distribution has the opportunity to determine which products are seen.

But AI Agents may transition this competition from “distribution” to “fund routing.”

Previously, you would ask: Which bank do I prefer?

In the future, a more important question may be: Which banks does my financial Agent include in the acceptable counterparty list?

Once, fund companies sought to ensure their money market funds appeared on app homepages.

In the future, Agents may directly determine: when the yield difference exceeds 18 basis points, liquidity requirements are met, and counterparty risk is below a certain threshold, where funds should automatically go.

Previously, lending institutions relied on advertisements, branding, and sales channels.

In the future, Agents may directly compare actual interest rates, collateral requirements, prepayment conditions, and the current cash flow of businesses, deciding whose funds should be utilized.

At this point, what truly matters is no longer just “do users have your app downloaded,” but whether you are included in the machine's default rules.

Whoever is written into the authorization scope qualifies to participate in competition; whoever ultimately secures fund routing captures the flow of capital.

Agentic Commerce alters the direction of demand flow.

Agentic Finance modifies the direction of capital flow.

So if today I were a bank CEO, I would of course care whether AI can reduce the time analysts spend on reports, whether it can automate KYC and anti-fraud investigations—these are very realistic efficiency improvements.

However, a more long-term question may be another:

Once my corporate clients have a financial Agent working 24/7, why should I continue to hold their next dollar?

The smartest banks may not resist this change; rather, they might be the earliest to integrate Agents into their systems. Clients’ operational funds would remain in checking accounts, temporarily idle funds would automatically enter higher-yield deposits or funds, and would automatically return when liquidity was needed; foreign exchange, credit limits, collateral, and payments could all be optimized together. This way, even if Agents continually seek better solutions, capital can still remain within the same banking ecosystem.

From this perspective, explorations like Project Agorá may not weaken banks; indeed, they could help regulated commercial banks maintain a core position in the stablecoin era.

Thus, this is not a simple story of “AI and Crypto will eventually eliminate banks.” The real threat may not be the banks themselves, but rather those financial business models that have long regarded customer reluctance to compare, switch, or find it tedious as their competitive moat.

In the past, a significant advantage for a bank was that moving clients was too cumbersome. The next strong banks will likely need to prove an entirely contrary proposition: clients can leave anytime, but after their AI has finished computing, they still choose to keep their money here.

In the past decade or so, the fiercest battles in Fintech have played out on mobile screens. Everyone competed for daily active users, main accounts, the most commonly used cards, and who could get closest to users.

The next round of truly important financial wars may not even have an interface.

It occurs within a set of authorization rules in the background, within an acceptable counterparty list, within yield differences of mere basis points, and at the moment when AI determines “where this money should go at this moment.”

Thus, the truly worthwhile question regarding Agentic Finance may never be: When will AI manage our money for us? But rather, as more and more money begins to seek its best destination according to rules machines can understand automatically—

who still has the authority to decide where this money ultimately stays?

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。