Anthropic launches AI agents for executing financial service tasks, what are they disrupting?

Writing Article: Web4 Research Center

Anthropic, the company that has swept the developer community in recent years with AI programming tools, announced on May 5th, U.S. time, the launch of 10 AI Agents specifically designed for financial services, officially initiating an assault on Wall Street.

According to Sina Finance, these 10 tools’ task lists nearly cover the core areas of daily work for financial professionals: preparing client meeting presentations, reviewing financial statements, and escalating cases for compliance review. The target users span professionals in banking, insurance, asset management, and fintech. This is not a chatbot, nor is it an auxiliary Q&A tool. It is a set of digital employees capable of directly embedding into financial institutions’ workflows and handling specific tasks.

Market volatility signals are far more complex than headlines suggest. Investors’ foot votes reveal a deep industry consensus: as AI Agents begin taking over tasks once considered irreplaceable for finance professionals, the entire value chain of financial services may be at a turning point.

1. From coding to reporting: a similar business logic path

Anthropic’s entry into finance mirrors its conquest of the programming market. To be precise, it’s almost the same script playing out across different industries. Before launching its financial AI Agents, Anthropic had already established a dominant position in the programming tools market. According to a research report published by Zheshang Securities in April 2026, Anthropic’s Claude Code held a 54% market share in enterprise-level coding Agents. By February 2026, 4% of public commits on GitHub were made by Claude Code, with analysts expecting this to exceed 20% by the end of 2026. In enterprise large language model spending, Anthropic accounted for 40% of the market, with 80% of the top global wealth companies being paying clients.

Data shows that Anthropic’s overall share of the US AI market has surged to nearly 70%, significantly eating into ChatGPT’s previous 90% share. From a challenger to a market leader in less than a year. The logic of disruption in the programming market is simple: AI Agents don’t just help programmers type more efficiently; they generate code, debug, deploy, compressing days of development work into hours. According to a survey by Economic Observer from October 2025 to January 2026 of 201 financial service practitioners in Mainland China and Hong Kong, 81% of financial firms have integrated AI into their workflows, but pain points remain—talent shortages, outdated systems, lagging regulation. These pain points are precisely the leverage points AI Agents can exploit.

But there’s a subtlety worth noting. Nicholas Lin, head of financial services at Anthropic, made a seemingly understated but profound statement. According to Tencent News, he said AI applications in finance are “only a few months behind programming applications,” which have already accelerated significantly. A few months’ difference—not years, not a generation of technology cycles, but just a few months. Behind this judgment lies a deeper logic: if the demand structure for AI Agents in finance is fundamentally similar to that in programming, then the dominoes of disruption in the programming market falling over will only be a matter of time for finance.

From concrete work scenarios, these 10 agents are assigned to two categories of tasks: five for financial research and client coverage, five for finance and operations. In research and client service, Claude agents can set goals, perform comparable company analyses, draft presentation materials, and prepare background summaries of clients and counterparties before calls. In finance and operations, they can check if valuations meet comparable company metrics, execute closing checklists, prepare journal entries, and generate closing reports. TechOrange’s report revealed more details: Claude can now run directly within Excel, PowerPoint, Word, and Outlook via plugins, meaning financial analysts don’t need to leave their daily software—AI Agents are embedded right there.

However, when AI Agents are embedded deeply enough, a more fundamental question emerges: if these Agents are not just drafting memos but start making financial decisions on behalf of institutions or clients, how far can their “hands” reach?

2. Same battlefield, two attack routes

Anthropic is not Wall Street’s only “door-opener.” Almost simultaneously, OpenAI launched its own financial push. According to Bloomberg Law on May 5, 2026, OpenAI and PwC announced a joint development of AI Agents targeting CFO teams, covering planning, forecasting, reporting, procurement, payments, finance, tax, and settlement processes. Interestingly, OpenAI positions its finance team as “Client Zero”—testing a procurement agent tool within its own financial operations before replicating the experience to enterprise clients.

Looking back, on March 6, 2026, Zhihui Finance reported that OpenAI released GPT-5.4, along with a set of financial service tools that connect to data sources like FactSet and Third Bridge, and can directly create and verify financial models in Excel and Google Sheets. On April 14, Wedbush published a research report revealing that OpenAI had officially acquired Hiro Finance, a startup focused on autonomous personal finance.

The paths of the two companies are becoming increasingly clear. Anthropic chose a bottom-up approach: starting from analysts’ workbenches, focusing on repetitive tasks that consume large amounts of manpower daily, gradually infiltrating into the operational systems of financial institutions. OpenAI, on the other hand, partners with consulting giants like PwC, pushing top-down from CFO offices, focusing on core control points in financial management. One route targets “efficiency gaps,” the other aims at “control heights.”

This speed is intriguing. It’s not a gradual, multi-year infiltration but a market encirclement completed within months. When the largest financial institutions start defining AI Agents as “digital colleagues” rather than “efficiency tools,” the shift in language reflects a deeper identity confirmation—these Agents are transitioning from “assistive tools” to “semi-autonomous participants.”

From assistance to participation, each step seems smooth. But moving from participation to autonomy requires a completely different infrastructure. An Agent that screens comparable companies for analysts and one that holds assets and executes payments for clients are facing nearly different technological challenges.

3. The gamble behind the gamble: who’s really getting the money?

The market responds to the advent of AI Agents with falling stock prices; another, more primitive way, expresses its belief: money. The timeline need not be long. In February 2026, Anthropic raised $30 billion at a valuation of $380 billion. Just two months later, Bloomberg and CNBC reported on April 29, 2026, that Anthropic was negotiating a new funding round of about $50 billion, aiming for a valuation of $900 billion. If successful, this would surpass OpenAI’s $852 billion valuation at the end of March, making Anthropic the world’s highest-valued AI startup.

In two months, valuation jumped from $380 billion to $900 billion—a magnitude rarely seen in business history. But more noteworthy is the direct catalyst behind this funding: Anthropic’s release of the Claude Mythos Preview model in April. This advanced cybersecurity-capable model was limited to about 50 institutions including Apple and Microsoft, sparking high-level meetings in Washington and Wall Street. Just a preview version, yet it propelled the valuation by hundreds of billions, fundamentally changing the market’s pricing logic for “trustworthy vertical industry AI.”

Capital bets are not only reflected in valuation. According to IT Home on April 30, 2026, Anthropic’s annual recurring revenue reached $30 billion, up from about $10 billion a year earlier—a nearly vertical growth curve.

Meanwhile, on May 5, at a company event in New York, Anthropic CEO Dario Amodei discussed AI alongside JPMorgan Chase CEO Jamie Dimon. The audience included Wall Street bank executives. A Silicon Valley founder stood in the spotlight of Wall Street’s power center. What are the bankers thinking? Perhaps not hard to guess. When asked about the surge in AI infrastructure spending, Dimon replied: “Overall, it makes sense. Trying to pick winners and losers would be very difficult.” This seemingly plain comment reflects industry anxiety—not about not wanting to choose sides, but about not daring to gamble wrongly.

Yet, a recurring question remains unresolved: if AI Agents are no longer just “digital colleagues” but need to directly hold assets, authorize expenditures, and sign contracts, how far can the existing financial infrastructure support them?

4. When Agents are more than “report-writing assistants”

This is not science fiction; it’s already outside the door. In May 2026, Odin Group officially launched OwlPay Agent Wallet, a digital wallet designed specifically for AI Agents. According to China Times on May 5, this is not a traditional digital wallet—its users are not humans but AI. After authorization, AI can send, receive, and manage stablecoins without manual operation. The wallet adopts a self-custody architecture, with users fully controlling private keys and funds, and all credentials generated and stored locally, supporting mainstream blockchains like Ethereum, Stellar, and Solana.

On the same day, GlobeNewswire also published related reports. Odin Group stated that the wallet leverages the company’s payment licenses in 40 US states, extending regulated stablecoin access to the AI Agent economy. This is not a proof of concept; it’s a fully operational product with compliance in 40 states.

So, why does a wallet designed for AI Agents need stablecoins and blockchain? Can’t AI Agents just use bank cards? Of course they can. As an analyst observed in late April: if an AI Agent is just helping a user buy a plane ticket, book a hotel, or renew a SaaS subscription, it can call existing payment systems like Swift, credit cards, virtual cards—no fundamental obstacle. But the real challenge arises in scenarios where an AI research Agent needs to access multiple databases, purchase paid resources, access different model APIs, pay for chart-generation tools, or even buy analysis segments from another Agent. In these operations, there may be no traditional storefront or checkout page. The Agent faces a series of APIs, data interfaces, model services, and computing nodes.

When the transaction entity becomes a machine, the traditional financial system realizes it is missing a crucial piece at the bottom. From a macro perspective, this is not an isolated business observation. AI Agents are evolving at a pace far beyond other infrastructure, transforming from auxiliary tools into genuine economic participants. Although Agents can now perform tasks and trades, they still lack standard ways to prove “who I am,” “what I am authorized to do,” and “how to get paid” across environments. “Identity is non-transferable, payments are not yet default programmable, and collaboration remains siloed.” Blockchain, as a public ledger, portable wallet, and programmable settlement layer, is being viewed by some tech teams as the key infrastructure to fill these gaps.

This is not wishful thinking about blockchain. As PwC pointed out in a report early 2026, financial institutions are gradually positioning AI as a “strategic transformation engine” rather than just an efficiency tool. When Agents evolve from “helping you do things” to “managing assets for you,” “verifiable execution records” become a survival threshold rather than a bonus—on-chain records are not meant to replace traditional audits but to provide a trustworthy trail at the Agent level that human auditors cannot monitor in real time. This means that in future financial ecosystems, Agents will likely need both traditional compliance channels and auditable on-chain identities and payment infrastructure—two parallel tracks.

However, it must be admitted: although OwlPay Agent Wallet has obtained payment licenses in 40 states, its overall adoption remains early; protocols like x402 and various Agent identity proposals are still in standard discussion; the concept of “Know Your Agent” (KYA), while gaining attention, is still far from widespread implementation. This is not a story already completed but one stumbling forward. Its value lies not in proving some irrefutable conclusion but in exposing a real problem: in the closed loop of traditional finance, machines are always tools, not subjects. But today, they are learning to do more.

5. Reinterpreting the anchor of value

This may sound like a picture of AI replacing human work. But if you pause to think, the real change might occur on a different dimension. The core value of traditional financial information services is built on information asymmetry. FactSet and Morningstar’s value lies not only in their data but in how they organize it into formats that professionals can call, compare, and model. This “organization cost” forms their moat. AI Agents’ logic, however, is entirely different. They are not organizing data; they are executing processes—they are operators, not repositories.

This distinction is crucial. FactSet’s stock price dropped 8.1% after the announcement, Morningstar fell over 3%, according to Sina Finance citing Eastmoney. But the reason isn’t just “AI can replace human analysts”—it’s a market re-pricing: when AI systems can connect directly to FactSet and Morningstar data sources for real-time analysis, data services themselves become raw materials, not finished products. Raw materials are always priced lower than finished goods.

This also explains why Anthropic, upon launching its financial AI Agents, announced a $1.5 billion joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs, aiming to accelerate Claude’s AI capabilities into more enterprise scenarios. At the same time, Claude can now connect directly to FactSet, S&P Capital IQ, MSCI, PitchBook, Morningstar, and deeply integrate with credit rating and corporate data sources like Dun & Bradstreet and Moody’s—over 600 million public and private company records flowing directly through Claude. The deeper implication: Claude isn’t competing as a data vendor but is redefining the decision-making layer above data vendors.

However, when information processing and decision-making are compressed into a continuous automation flow, the human roles along the chain face redefinition. From briefings to compliance upgrades, these 10 AI Agents are inserting into three previously considered “irreplaceable” steps: information organization, professional judgment, and risk control. Each is being partially disassembled. Analysts no longer monopolize information organization; compliance teams no longer monopolize initial risk screening; investment banking VPs no longer monopolize pitch material writing.

This doesn’t mean “people” will be completely replaced. But it does mean that human roles are shifting from process operators to process designers and supervisors. This isn’t just “job anxiety.” It’s more like a river changing course: the water volume remains, but the riverbed shifts. Old docks may be abandoned, new docks are hurriedly built downstream.

This evokes a classic philosophical metaphor. Heidegger, when exploring technology, was concerned not with tools themselves but with technology as a “standing-reserve,” reorganizing our relationship with the world, changing how we see things, others, and ourselves. The embedding of AI Agents into financial workflows is weaving a new “standing-reserve.” This new framework not only handles data and reports but also redefines what the core value of financial work is.

6. Not the end, but a watershed

FactSet, Morningstar, S&P Global, and Moody’s stocks suffered heavy hits, signaling market implications. According to Eastmoney, FactSet once dropped 8.1%, Morningstar over 3%. On Wall Street, such figures mean the market is betting real money on a judgment—traditional financial info providers’ moat appears more fragile in front of AI Agents than previously imagined. But fragility doesn’t necessarily mean immediate disappearance. A different evolution may occur: a restructuring of the value chain. FactSet and Morningstar possess irreplaceable data assets, the fuel AI Agents rely on. The question is, when the fuel itself is no longer scarce, the scarce resource becomes the engine that injects fuel precisely into the combustion chamber. The engine makers are taking a larger share of the value chain.

A noteworthy detail: according to Zheshang Securities’ April 2026 report, one key to Anthropic’s success is its focus on an auditable rule-based framework. Compared to competitors like OpenAI and Google, Anthropic emphasizes traceable reasoning and transparent compliance systems, making it naturally suited for high-regulation industries like finance, law, and government. In a trust-based domain like finance, the safety and compliance positioning of AI companies may be more durable than model capabilities. It’s not about who is smarter but who is more trustworthy. The latter carries far more weight on Wall Street.

AI Agents are evolving from coding tools into active participants in the real economy. As they begin to play roles beyond tools—owning assets, authorizing expenditures, signing contracts—the language of financial infrastructure is being rewritten. Payments, identity, rights and responsibilities, audits—these foundational concepts of modern finance face redefinition when confronted with “invisible participants.” This redefinition occurs within traditional finance but also spills outside, prompting new infrastructure explorations.

Wall Street’s financial AI Agents are just the beginning. When Goldman Sachs and JPMorgan deploy Agents into core workflows, when FactSet and Morningstar redefine their value propositions, and when OwlPay creates dedicated wallets for Agents—these seemingly isolated events are actually piecing together a larger picture: Agents are no longer just “doing things for people,” they are beginning to participate in value distribution.

The final word is a response to this vision: Agents are in the game, and the rules are just beginning to be written.