Major domestic tech companies have been aggressively rolling out AI-powered office tools this year, with an array of applications such as WorkBuddy and Qianwen Office launching in rapid succession. AI agents are consistently delivering value in high-frequency office tasks like code writing, PPT creation, document organization, and data analysis, as the office setting naturally bridges the consumer and enterprise sides, offering frequent demand and tangible user experience. This makes it a prime gateway for the AI boom.

Beyond the buzz surrounding "office productivity," a deeper battleground has already opened up—AI is beginning to move from the desk to the core business operations of enterprises. This trend is visible both at home and abroad. In overseas markets, Palantir has continued to attract strong capital market interest through its deep application of AI in decision-making, operations, and complex business processes, with its market value surpassing $400 billion. In the domestic market, AI-to-B has become a key direction for many large model companies, with Zhipu at one point reaching a market valuation of nearly one trillion Hong Kong dollars, fueling rising commercial expectations in the industry.

However, getting AI into core business processes is far more difficult than in office scenarios. According to an MIT study, 95% of enterprise AI pilot projects fail to successfully deploy, primarily because the tools cannot learn, integration is poor, or they do not align with actual workflows. This means enterprises are never truly focused on single-point efficiency gains, but rather on whether AI can integrate into real business chains and directly support revenue growth, user conversion, and transaction fulfillment. In the era of the agent economy, whoever can help enterprises cross the critical line from pilot to scale, enabling AI to truly generate income, carry business, and complete transactions, is more likely to hold the core leverage for the next decade.

Where to Begin: Beyond the Office

Office work is the first stop for agents to reach non-technical users. Tasks like coding, making presentations, organizing files, and analyzing data share common traits: clear objectives, digitized processes, and easily measurable results. In these tasks, agents solve efficiency problems. But in real business environments, the core value enterprises need often doesn't happen at the desk. Customer acquisition, marketing, store operations, customer service, inventory, transactions, and fulfillment are where merchants generate actual business results every day. This means that in commercial scenarios, the challenge for agents is whether they can connect multiple business segments and achieve closed-loop delivery—a path that extends agent applications into business operations.

Just as mobile internet reorganized the connection between people and services, the future agent economy will shift users from "finding and operating services" to "stating needs, letting agents identify intent, plan, and execute." AI agents will become new service entry points and business models, reshaping the entire digital economy. Domestically, Alipay launched its AI life assistant, Abao, in June, which has already completed AI-based integration for over 10,000 services. Through cross-device and multi-entry points, users can call on intelligent services from different terminals like car systems and smartphones. For the tens of millions of merchants in the Alipay ecosystem, this means business previously scattered across various apps, stores, and service systems has the opportunity to be reorganized by a single intelligent gateway connected to over a billion users.

This path isn't exclusive to domestic players. OpenAI is also positioning AI for commerce; ChatGPT has supported product discovery and purchasing since last year, with agreements to connect agents with merchants and payments. Recently, consumer finance company Synchrony announced an enterprise partnership with OpenAI, allowing consumers to use a dedicated credit card within ChatGPT to purchase goods, integrating lending and credit into an AI-native shopping and checkout experience. Behind this lies a new market space; McKinsey predicts that by 2030, AI agents will participate in $3 trillion to $5 trillion in global consumer transactions (Agentic Commerce), making the deep integration of agents and consumption a clear industry trend.

In other words, the vast future agent market won't come solely from office efficiency; commercial scenarios will also be a key focus. As users get used to handing needs directly to agents, AI, beyond being a tool, could become a new gateway connecting users to commercial services. The deep integration of consumption and agents also raises new demands for merchants in this era: when users start to search, compare, call, and purchase services through agents, can their businesses enter this new interaction system? In the past, merchants needed to make themselves easy to find and recommend via search engines; in the future, they must also make their businesses understandable and callable by agents, and further complete services and transactions.

From "Integration" to "Delivery": Hurdles for Intelligent Operations

For agents to truly enter the operational scene, the first problem is how to AI-enable existing mini-programs and APIs, quickly integrating them into the agent ecosystem. For many small and medium-sized enterprises, this traditional approach demands significant time and technical resources, creating the first barrier to adopting AI agents. At the Alipay AI Ecosystem Conference held on August 17, Ant Digital Technologies launched Agentar Ecosystem Edition. Leveraging a decade of ToB accumulation, it provides merchants with one-stop AI capabilities from integration and operations to custom development, allowing business capabilities scattered across different systems to smoothly enter the agent ecosystem.

For example, a merchant can simply say "I want to connect to Abao," and their existing mini-program or services are quickly transformed into standard Skills callable by Abao. After previewing and confirming, they can publish with one click. Services that once existed in business systems become capabilities that agents can understand and invoke, reaching over a billion consumers through diverse terminals like phones, car systems, and smart glasses. Integration is just the first step; the real challenge for agents entering the operational scene is connecting real business processes and delivering stable results. Writing a marketing plan is a one-time output for an office agent, but in real business, the plan must be executed, sales data must translate into concrete operational practices, customer service replies must actually complete the service, and inventory forecasts must coordinate with procurement and sales.

Addressing merchants' operational pain points, Agentar Ecosystem Edition has now accumulated over 200 Skills for high-frequency merchant scenarios, covering four categories: growth, operations, experience, and general capabilities. These include lead mining, sales forecasting, intelligent data querying, inventory alerts, and intelligent customer service. Merchants can directly invoke and combine these standardized capabilities, turning AI from a one-off tool into a continuously callable digital employee. For instance, a merchant asking "help me analyze this week's store performance and give marketing suggestions for next week" requires reading operational data, analyzing sales, identifying anomalies, then invoking marketing capabilities to form a final strategy. Agentar provides role-level digital expert agents, like the "AI Store Manager," which can autonomously plan paths and dynamically schedule multiple capabilities upon receiving a business goal, ultimately delivering operational analysis or marketing strategies.

For merchants with highly customized needs, Agentar packages agent construction into standard templates, allowing secondary development through visual drag-and-drop or one-sentence programming to create bespoke agents. Agentar also offers engineering services such as consulting, custom development, system integration, and ongoing maintenance, with FDEs (Frontline Deployment Engineers) supporting merchants in private deployment and continuous optimization. This means agent capabilities are moving from "completing tasks" to "carrying business," and will increasingly be implemented in the real operational scenarios of more SMEs. For instance, Chongyu Technology used Agentar to create a "digital clone" of pet doctors, supporting multimodal consultations via images and voice with 24/7 presence, adding 18-25 new customer leads per hospital monthly and saving doctors about one-third of their time. This shows how professional services once heavily dependent on individual expertise can become callable AI service capabilities. Similarly, Yiqilian Technology uses AI badges to record customer needs, with six business agents covering acquisition, conversion, delivery, and after-sales. They automatically extract over 20 structured fields and score service quality, boosting new customer conversion and membership renewal rates by 10%-20%. Both cases validate whether operational segments previously reliant on manual experience can be broken down, integrated, executed by agents, and ultimately converted into measurable business results. Throughout this process, new knowledge and experience generated during operations can be further distilled into Skills via Agentar, creating reusable, iterable organizational assets.

From "Usable" to "Reliable": Ant Digital Builds Agent Infrastructure

When agents overcome numerous hurdles to enter the business arena, they face more practical issues: can they run long-term, stably, and at scale once in the business? The first issue is cost; when agents shift from occasional use to high-frequency operation, token costs directly become a major component of business operating expenses. Next are security, controllability, and stability. The closer agents get to transactions, services, and fulfillment, the higher the demands for accuracy, stability, and security. When agents truly enter business, they involve permissions, data, knowledge, call chains, and cost control, making security, evaluation, and governance fundamental capabilities in production environments.

Ant Digital's Agentar, built on the LingDT foundation model and industry-specific large models, doesn't compete on parameter count but on "intelligence-to-efficiency ratio"—achieving the same task with lower token consumption, helping enterprises spend every token wisely. Additionally, Agentar offers three modes—no-code, low-code, and high-code—covering the full lifecycle of development, evaluation, and operation, and supports data governance, knowledge engineering, and security compliance, embedding security and compliance as native capabilities. These capabilities have already been validated in large-scale practices across finance and energy sectors. The financial industry, with its stringent security and compliance requirements, serves as the most demanding testing ground. Agentar has deployed over 300 specialized agents in banking, securities, and insurance scenarios. A leading joint-stock bank uses agents across its entire customer operations process, improving end-to-end efficiency for relationship managers by dozens of times; the world's first AI-native bank, Ryt Bank, surpassed 1.2 million users within seven months of launch, processing over 25 million transactions. This capability has also been validated by third-party market reports. According to IDC reports released in 2026, "China Agent Development Platform Market Share, 2025" and "China Financial Industry Large Model Platform, Agent Applications and Services Market Share, 2025," Agentar ranks first among non-cloud vendors in the private market for agent development platforms and holds the top position in financial AI, large model, and agent service market share. For agents, moving from "being able to complete tasks" to "being able to enter production" is not just about model capabilities, but a full set of cost, engineering, security, and business deployment abilities—this is precisely where Agentar's advantage lies.

Final Thoughts: The Real Test Begins When Agents Enter Business

Office work will remain an important scenario for large-scale agent deployment. However, when agents enter segments like operations, services, transactions, and fulfillment, the standard for measuring them will shift to whether they can carry business and ultimately deliver results. From this perspective, Ant Digital's Agentar is attempting to leverage years of industrial digitalization experience to transform merchants' knowledge, experience, and processes into capabilities that agents can invoke, combine, and execute. By tapping into Alipay's user and commercial ecosystem, it aims to push agents further toward actual delivery. For Agentar, the next phase worth watching will be how it replicates its proven industrial capabilities across more industries and merchants, while consistently delivering measurable business outcomes.