Just last week, Tencent officially launched the WorkBuddy Finance edition, offering over 80 financial experts and expert panels to banks, brokerages, and insurance companies.

Less than a month ago, Baidu renamed GenFlow to "Kuku AI," and Finance was chosen as its first key scenario for transitioning from general office to professional office. Going further back, ByteDance's subsidiary Kouzi has brought in Huatai Securities, GF Securities, and Guosen Securities, bringing market data, financial data, and ETF screening capabilities into the Agent platform.

From ByteDance and Baidu to Tencent, the competition among general agents is rapidly expanding into the financial sector.

But here's the question: is it really worth making a separate Agent version of the financial market? Even the big companies themselves haven't figured it out.

In today's article, Kingtech will thoroughly explain the underlying logic, competitive landscape, and investment opportunities of this "financial agent battle" all at once.

|Why finance? Because the "hard labor" here is the most valuable

The logic behind big companies choosing finance is actually quite simple—the "hard labor" in the financial industry is too expensive.

Positions like investment banking, research and investment, credit, and insurance have extremely high unit value per hour, but much of the time is spent searching for data, flipping announcements, verifying calibers, updating models, and preparing materials. A single corporate credit due diligence process takes about 10 days; Multi-channel public opinion review, starting at 2-3 hours; A fund for in-depth research usually takes 1-2 days.

As long as the agent can take over some of it, it has the chance to directly convert it into efficiency that financial institutions are willing to pay. The actual data provided by WorkBuddy Finance Edition shows: credit due diligence has been compressed from 10 days to 1 day, public opinion review from 2-3 hours to 10 minutes, and fund in-depth research from 1-2 days to 30 minutes—overall efficiency improvement about tenfold.

More importantly, much of the financial work is built on digital data—market data, financial reports, announcements, research reports, and industrial and judicial information mostly exist as structured data or electronic documents. This means Agents don't need to "understand the world from scratch"—they only need to connect existing digital assets.

In short: Finance is one of the easiest industries to settle business accounts in agent implementation.

|The battle of three routes: C-end personal version vs. B-end industry version vs. general platform version

Although ByteDance, Baidu, and Tencent have all expanded their agent services into finance, they are not targeting the same market.

ByteDance Button—C-end "Skill Store" model

Kouzi's approach is to break down professional financial capabilities into individual skills and directly integrate them into the agent platform. Currently, the Kouzi Skills Store has launched a series of financial skills from institutions such as GF Securities and Guosen Securities, covering high-frequency research scenarios such as financial comparison, dragon-tiger rankings, ETF screening, ETF fund movements, and regular fund investments.

Users only need to tell Kouzi what they want to research, and it can use the corresponding skills to query market data, compare companies, and filter ETFs. Multiple skills can also be combined into continuous tasks for pre-market information organization, intraday monitoring, and post-market review.

Baidu Kuku AI—the 'content + storage' model for consumers

Kuku AI relies on Baidu Wenku's 1.8 billion professional documents, Baidu Scholar's 700 million documents, and Baidu Netdisk storage capabilities, and includes financial data such as listed company profiles, stock quotes, financial reports, and research reports.

After users request research tasks, Kuku AI can perform data retrieval, data processing, and content generation, ultimately delivering Word research reports, financial analysis PPTs, or financial model Excel directly. Long-term monitoring tasks can also remain in the cloud to continue execution. Kuku's AI office MAU has surpassed 25 million, ranking first in the general AI office sector.

Tencent WorkBuddy Finance Edition—B-end "industry-exclusive workbench" model

Tencent is taking a completely different path—launching independent financial products directly for banks, brokerages, insurance companies, and other financial institutions. Focusing on the four main working platforms of "corporate finance, retail finance, investment research and consulting, and individual customer operations," more than 80 dedicated financial experts and expert teams have been launched.

In corporate credit due diligence, it can automatically generate material lists, use industrial and commercial financial and judicial data for cross-verification of risks, and output initial due diligence drafts according to internal bank templates; In investment research scenarios, integrating Morningstar fund data supports fund screening and portfolio diagnosis; In insurance scenarios, it covers the entire process from customer acquisition to order tracking.

According to Tencent, since March this year, WorkBuddy has successively entered more than 100 financial institutions, including CICC, SDIC Securities, Ping An Bank, and China Taiping. However, according to feedback from multiple securities firms, it has not yet been widely adopted internally.

| Core Divergence: Is a Universal Agent Sufficient, or Must We "Redo the Whole"?

Behind these three routes lies a deeper strategic divergence: Should general-purpose agents continuously expand their capabilities through data connectivity, knowledge bases, and skills, or should they rebuild deeper products for high-value industries like finance, law, and healthcare?

Baidu insiders put it bluntly: Kuku AI is still positioned as a general office agent, while finance is more seen as a "sample" demonstrating complex task capabilities. If an agent can handle complex tasks like finance, it can intuitively demonstrate the capability ceiling of a general agent.

Tencent has chosen a more aggressive path—launching a financial edition, putting banks, brokerages, and insurance companies into a separate product. This means Tencent is gambling: financial institutions are willing to pay extra for a specialized industry agent, and office agents have the chance to evolve from "selling general productivity tools" to "industry-based" pricing.

Overseas giants have also failed to reach a consensus answer.

OpenAI and Anthropic lean more toward a "general agent + financial plugin" approach—OpenAI introduced ChatGPT into the investment banking scenario through the Investment Banking plugin, and Anthropic launched 10 sets of financial agent templates. Google went straight to the industry version, launching Gemini Enterprise for Financial Services in August this year, which includes over 50 financial skills.

| Three hard hurdles: data, security, and commercialization

Financial Agents sound impressive, but there are at least three rigid hurdles in the implementation process.

Threshold One: Data accuracy and traceability

Every conclusion in the financial industry must stand up to scrutiny. Inconsistent data standards, missing sources, and untraceable conclusions make professionalism and compliance impossible to talk about. WorkBuddy Financial Edition integrates multi-source data such as private fund performance, analyst consensus expectations, real-time market data, business information verification, and global fund ratings, and requires conclusions to retain sources. However, the performance of a generic agent on the open network may not remain reliable once it enters the financial business field.

Threshold 2: Security compliance and permission governance

In June this year, the National Financial Regulatory Administration issued the "Guiding Opinions on the Safe Development and Application of AI in the Banking and Insurance Industries," listing fund transactions, asset evaluation, credit approval, underwriting and claims processing, and risk management as high-risk AI applications, requiring the establishment of manual supervision and intervention mechanisms at key stages, and the retention of original data and reasoning path records. WorkBuddy Finance uses a "isolated computing power + dedicated network + local security governance" architecture, offering three solutions: private deployment, dedicated VPC, and SaaS, deeply integrated with WeChat Work, with key actions requiring manual confirmation.

Threshold 3: The commercialization path is still unclear

Dongwu Securities clearly pointed out in its research report that most agent products are still in the free trial stage, and the paid conversion path remains unclear. Moreover, fluctuations in the equity market directly affect C-end usage frequency and B-end willingness to pay.

Jingtai Perspective | AI Office Platform—From "Entry Battle" to "Ecosystem Positioning"

According to IDC data, the market size for enterprise-level AI Agents in China will be about 21.2 billion yuan by 2025, expected to grow to 44.9 billion yuan in 2026, and exceed 332 billion yuan by 2029. As of July 2026, Tencent WorkBuddy has 6.582 million monthly active users, ByteDance TRAE Work 1.904 million, and Alibaba QoderWork 230,000. Tencent currently leads in the AI office track, but ByteDance's growth is even fiercer.

Focus areas: Tencent (WorkBuddy ecosystem), Baidu (Kuku AI user base), Kingsoft Office (WPS AI native product "Lingxi").

Financial data vendors—both beneficiaries and disruptors

The massive query demands of financial agents have opened new commercialization paths for data vendors, but MCP standardization may also weaken user stickiness in traditional terminals. Data owners like Tonghuashun and Eastmoney are facing both the risks and opportunities of being "platformized."

Focus areas: iFinD (iFinD data + MCP opening), Eastmoney (Choice data + Miaoxiang AI), Hundsun Technologies (financial IT infrastructure).

Overseas benchmarking—Rogo model validates independent valuations of financial agents

The valuation of financial AI company Rogo soared from $350 million in April 2025 to about $2 billion in 2026, nearly sixfold in one year. Its core problem is to build a "translation layer" between foundational models and financial work—connecting downward to specialized databases like Capital IQ and FactSet, generating directly usable Excel models, PPTs, and research reports upward, while retaining data sources and original citations. Currently, Rogo serves over 300 institutions and more than 40,000 financial professionals.

This validates a judgment: financial agents have the potential to evolve from "ancillary functions of general tools" into independent software categories.

Actual efficiency improvement data comes from vendor self-reports, and current applications are still concentrated in a small range of financial institutions. Agents' autonomous execution capabilities remain limited, and the implementation of complex financial task scenarios will require time.

Most financial agents are still in the free trial phase. If financial institutions ultimately refuse to pay extra for the "industry edition," then general-purpose agents relying on stronger models and richer skills can cover most demands, potentially squeezing the independent space for the "industry edition."

For investors, the most important thing right now is not to bet on a single company, but to focus on a core indicator: are financial institutions willing to pay for agents? If the answer is "willing," financial agents will become an independent software category, and the entire high-value industry will be re-segmented; If the answer is "wait a little longer," then this battle will take a long time.