Alibaba Cloud Bets on Financial Agents: Can Banks and Brokerages Welcome Digital Employees?

"Last year was AI Plus, this year is AI Native."

In just a year, in the financial sector—the track with the richest wealth imagination—AI and agents have been thrust into the spotlight again.

By 2025, large models have become a standard project for banks, brokers, and insurance companies, and Agent has become one of the hottest technical terms. But behind the buzz, a question remains unresolved: How much value have these Agents actually created?

In fact, in many financial institutions, so-called agents are more like add-ons. They can summarize materials, recognize images, answer questions, but find it hard to truly enter core business processes. Customer service is still customer service, client managers are still client managers, researchers still need to accomplish most of their work themselves, and AI acts more as an assistant.

This situation is quickly changing as agents iterate.

On June 16, 2026, Alibaba Cloud unveiled a new card in the financial sector at the China International Finance Expo—the "Dianjin" agent. Along with the product, a lengthy list was brought to the forefront, announcing the official opening of over a hundred professional skills in wealth management, credit risk control, insurance business development, investment analysis, and more, allowing financial institutions to directly utilize them in agent development.

According to the official description, Dianjin can now execute complete task chains in business processes such as investment research, advisory, credit, risk control, claims, including running backtests, training models, writing reports, doing compliance—a series of tasks previously thought to require professionals.

"This year is truly the inaugural year of intelligent agents," stated Zhang Chi, Vice President of Alibaba Cloud Intelligence Group and General Manager of the New Finance Industry, to Wallstreetcn.

In Zhang Chi's view, the financial sector is moving from AI Plus to AI Native, from enhancing business processes with AI capabilities to letting Agents directly enter specific job roles. From client manager, credit manager, insurance agent to broker researcher, everyone is beginning to have a digital assistant—perhaps even a digital avatar—by their side.

This may imply that large model applications in finance are bidding farewell to demonstration and experimental phases, gradually entering production systems.

AI is no longer just an institution add-on

Over the past year, anyone involved in AI projects within financial institutions has likely experienced an awkward scenario.

At the start of the year, dozens or hundreds of agents were promised in project setups, but at year-end reviews, most "agents" turned out to be mere add-ons to existing business processes. Processes, roles, and organizations remained unchanged; only an agent was added to some step.

In Zhang Chi’s opinion, many products called Agents in the past were essentially process orchestration systems. One customer service outbound call might relate to six processes, one credit business to over twenty, a claims process even to forty or fifty steps. AI was embedded at just a few nodes for recognition, generation, and summarization.

Zhang Chi used a metaphor: "We used to have a TV, we added a USB interface add-on to play things; today, it’s not that, the TV itself is a completely new species with AI fully integrated."

"Last year, AI was about post-training models to generate content for you. Today, it truly helps you understand the world and environment, serving as a genuine assistant, continually interactive." Zhang Chi believes financial institutions’ understanding of agents is shifting from AI Plus to AI Native.

The sign of the leap is the Agent sitting at a specific job position.

Zhang Chi cites client manager as an example. Previously, AI mainly helped at nodes in the investment research process, like summarizing reports, organizing materials, or generating content. Today’s agents can participate in the entire investment research chain.

"From industry analysis, selecting targets, uncovering undervalued companies, to generating trading strategies, modifying strategy codes—it’s a sustained workflow, more like a genuine AI assistant."

Alibaba Cloud’s new “Dianjin” universal financial agent platform repeatedly emphasizes digital employees.

CITIC Securities, following the grading logic of autonomous driving, divides digital employees into five levels: L1 to L5—from basic automation of single processes, to collaborative assistance, conditional autonomy, high autonomy, ultimately toward all-scenario ultimate intelligent agent. Behind this grading is the financial industry beginning to accept agents not as software functions, but as a new job type within organizations.

But why is Agent first landing in the financial industry? The answer relates to the sector’s own degree of digitalization.

Over the past five years, China’s financial sector has been one of the industries with the highest digital investments. Banks completed cloud-native renovations, brokers built data middle platforms, insurance companies pushed online operations. Huge amounts of business rules, customer data, and processes have been digitally deposited.

This is the foundation for Agents entering production systems.

Zhang Chi mentions an example: Guangdong has furniture, Henan has wigs, Jiangsu-Zhejiang has aquaculture. For regional bank client managers, these industries have entirely different business logics, financing needs, and risk structures. "This was the problem that digitalization couldn’t solve."

Traditional systems can record data, but can’t understand industries. Now, Agents are gaining this capability. They can combine industry chain info, order stats, exchange rate changes, and financial product design to offer tailored solutions to client managers.

Another change is happening in brokerage and wealth management. Here, large numbers of knowledge workers gather—reading reports, analyzing financial reports, building models, designing investment strategies daily. These tasks are naturally fit for Agent participation.

Behind these cases lies a bigger trend: The object of enterprise purchasing is changing, from SaaS software to digital labor. Zhang Chi repeatedly mentioned to Wallstreetcn concepts like "one assistant per post," "one avatar per post," "digital employee." The underlying logic highly aligns with the Agent economy depicted by Microsoft and OpenAI.

In the future, enterprises will measure productivity not just by employee numbers, but by the number of digital employees. Hence, Alibaba Cloud proposes a new metric—Token. "Digital employees cost more in tokens," Zhang Chi said. "It’s both a cost metric and a value metric."

Under this narrative, the financial sector is beginning to show new organizational forms: researchers with research agents by their sides, client managers with client manager agents, insurance agents with insurance agent agents. Gradually, a person owns a digital team.

The Full Stack Bet: Chip-Cloud-Model-Agent

Seeing the opportunity, Alibaba Cloud is gearing up to redefine its own role in finance.

Looking back over the past decade plus, cloud vendors’ stories in finance have been much the same: strategic co-builder, technical partner, long-term service provider. Building a data middle platform, digital transformation—projects start at six months or a year, by the time the system is done, business patience may be exhausted. Zhang Chi candidly says, "The original software and tech were too far from management and organization."

But agents are different—they can read a financial report, run a credit process, generate a research draft in a short time. "Today’s AI, why everyone uses it as a platform, is because it offers real-time feedback, real-time effect, and true business integration," Zhang Chi told Wallstreetcn.

This difference in pace offers Alibaba Cloud a window to reshape its positioning.

It tries to tell a new story—positioning itself as a full stack chip-cloud-model-agent AI infrastructure provider for finance: base layer is Pingtouge Zhenwu M890 AI chip; middle layer is Feitian Zhisuanyun (cloud) and QWEN model; upper layer is Dianjin agent platform and 129 open financial skills.

The purpose of this four-layer chain is to drive token unit costs to levels its rivals can't match.

Zhang Chi himself is a "modelist," but he’s not optimistic about the moat at the model layer. "Base models are open source, everyone is equal. Agent tech is also being done, there’ll be some uneven progress, because with AI Coding, leveling and complementing each other is very fast."

He thinks the real killer is data. "We chose data as a very big focus. Only when data delivers value does it become the key for agents in the financial industry. Last year, agents could speak well; this year, they can write and calculate—there’s a huge difference, since only data truly reflects the ultimate delivered result."

This also explains why Dianjin platform’s native integration of Wind, Yingmi Qieman, Eastmoney, Hangseng Juyuan is a major selling point, because he believes that what financial agents need to do is to generate strategy recommendations for portfolio managers based on real market data.

Alibaba Cloud is clear about its story’s boundaries: serving banks, brokers, insurers—licensed institutions—and first offering digital assistants for financial professionals.

Asked whether agents involve financial licensing, Zhang Chi is forthright: "We’re doing everything in line with regulatory requirements; focus is on person-serving, all products are ToB for financial institutions, licensed entities, not ToC."

Alibaba Cloud’s plan is clear—it wants to embed itself into the daily production flows of these institutions ahead of everyone, so that when the model layer is as competitive as possible and the engineering layer levels out, what remains is data, industry know-how, and client relationships.

After Agents enter deep waters

But now, the story must stop and look at the other side.

For financial agents to go deep, it’s not just about model capabilities and generation results—it’s about whether data, risk control, and compliance can keep up. These are the boundaries inherent in finance, and the hurdles Alibaba Cloud must surmount.

First is data. How to combine public and private institutional data is the first problem faced by all financial agent projects.

Zhang Chi’s solution is two-fold: "For public market information—including transaction data, and even licensing data from loans, customs, or the State Administration of Foreign Exchange, all can be used for research and analysis."

The private part must be localized. "For example, doing industry analysis—the public research has to be done partly, then after completing it, you need to use the internal investment research framework and knowledge base to do precise work, such as ETF or specific product design—you have to return to localization."

This approach works for leading institutions, but for small and medium-sized banks, localization is a significant investment—not to mention ongoing algorithm iteration costs. Alibaba Cloud’s solution is to use Agents for algorithm equality.

Zhang Chi explains, "Previously, regional banks lacked data algorithms and scientific talent, so in serving local loans, the service wasn’t standardized; serving so many enterprises and doing profiling, risk modeling—it was hard. Now, with Agent, past risk modeling logic can be fully used."

It sounds appealing, but for small and medium banks to treat Agent as a data scientist still requires a lengthy validation period.

Next is traceability and explainability.

The financial sector does not accept black-box processes; every tool invocation, every decision node, every manual intervention must be fully traceable. Dianjin claims to have credential isolation, operation traceability, and triple compliance defenses—making the entire reasoning process explainable and auditable.

But to what granularity this mechanism meets regulatory requirements remains to be seen in each business line's real-world checks.

Today, most financial institutions demand agent traceability far exceeding traditional logs—they want answers to "why the agent decided this step," which is exactly what large models struggle to explain.

Finally, responsibility attribution.

When agents participate in loan approvals, investment recommendations, wealth allocation, insurance claims—core business lines of financial institutions—who is responsible when something goes wrong?

Take credit, for example. Dianjin has achieved nearly equal approval rates between expert manual review and Agent review at a Southeast China bank, but Zhang Chi especially stressed, "It’s not yet fully offloaded in the true sense—this is a big challenge. Why not offload? Because the responsibility issue isn’t solved."

Another repeatedly mentioned but easily overlooked pinch point is "Number One Cognition."

Zhang Chi says: The biggest challenge for agent implementation isn't technology. "It’s the client’s own cognition, the cognition of the Number One—the human leader—which determines what happens with the agent."

Because the institution’s internal structure, assessment mechanism, and cultural climate determine whether agents are adopted as innovation showpieces or genuine productive tools. CITIC Securities was able to roll out the L1-L5 digital employee system not just through tech selection, but through top-down organizational consensus.

Returning to the initial question: Is 2026 the inaugural year of financial intelligent agents? The answer doesn’t depend on whose model is stronger—Alibaba Cloud, Huawei, Tencent, ByteDance; nor whose agent platform is more complete. It depends on whether financial institutions can find a workable balance between innovation and governance, application and compliance.

Letting Agent sit at a workstation is one thing; making it able to truly sign and stamp documents is another.

Risk DisclaimerMarket involves risks; investment should be cautious. This article does not constitute individual investment advice nor considers any user’s specific investment goals, financial situation or needs. Users should consider if any opinions, views or conclusions in this article suit their specific circumstances. Investments based on this are at your own risk.