Factory model that puts intelligent agents "to work".
Author Tao Huidong
Word count 3770
Handcrafted quantity丨100% AI content丨0%
Whether it's trendy or genuinely useful, the large-scale "employment" of AI agents in businesses has become an unstoppable trend.
IDC predicts that by 2030, the number of active digital employees (Agents) worldwide will reach 2.2 billion; Gartner predicts that by 2028, the average Fortune 500 company will use more than 150,000 AI Agents.
Currently, implementing an intelligent agent is very easy. Just drag a low-code canvas, connect a large model, write a few lines of prompts, and a "digital employee" can be operational in a presentation.
However, the other side of the buzz is the lack of substance. Also according to Gartner's prediction, by the end of 2027, more than 40% of intelligent agent AI projects will be canceled due to rising costs, unclear business value, and insufficient risk control.
Many companies find that while the number of agents is increasing and each reports several times the efficiency improvement, the overall costs have not decreased, and performance has not increased; they've just added another line to the token bill.
Out-of-control army of Agents
Why are agents not as useful as many companies imagine?
Gartner states in their report that most agent projects are still merely "early experiments or proofs of concept driven by hype, and are often misapplied." There's a specific term called "agent washing": re-labeling AI assistants, RPA, and chatbots as agents.
Even if the agents are genuine, they often struggle to complete the last mile. A Deloitte survey shows that 66% of organizations are experimenting with AI agents, but only 11% have actually deployed them in production environments. Professor Fang Yue from the China Europe International Business School summarizes it in four words: "Pilot prosperity, value nullity."
There's something even trickier than "not being usable": a lack of control. With the uncontrolled expansion of agent numbers, no one can definitively say how many agents are operating within a company, what data they can access, and who is responsible if problems arise.
The problem is clear: building an agent is not enough. The vast majority of companies do not lack agents; they lack agents that truly understand the core business as well as the capability to "manage" those agents.
Recently, I heard a "joke" from an investor: a very popular general agent product started promoting its enterprise version, claiming to have many impressive features like organizational control, security compliance, and collaborative sedimentation, yet the actual result was poor sales because it simply did not understand business operations; many customers preferred to purchase individual accounts rather than the enterprise version. Consequently, the company began to push FDE hard.
But FDE faces almost as many issues as the agents themselves. Many so-called FDEs are merely traditional implementation teams donning a new label, with no essential difference from the already debunked project-based model by the previous "AI Four Little Dragons." Very few companies can create a compounding know-how flywheel like Palantir.
At the same time, some players are exploring a different path: the Agent Factory.
The idea is that since the fundamental carrier for AI to enter enterprises is agents, then helping companies continuously produce and operate an army of agents in a cost-effective and reliable manner is the core of realizing AI, and this can be factory-ized.
Factory model that puts intelligent agents "to work"
An intelligent agent factory essentially transforms the previous reliance on a single engineer's experience in building agents into standardized, reusable products and processes.
From domestic and international industry practices, the evolution towards a "factory model" is becoming a consensus among giants and platform players.
Microsoft’s Agent Factory integrates Copilot Studio and Microsoft Foundry into a unified procurement and billing solution; Oracle's Private Agent Factory provides pre-provisioned agents and visual orchestration tools centered around enterprise databases.
Each has a different entry angle, but the core aims to integrate agent development, data access, and operations management into one platform.
In China, Ant Group's Ant Tech proposed the "Super Factory of Agents" enterprise AI infrastructure solution at WAIC this July, with the commercial intelligent agent full-stack platform Agentar as its core.
Compared to past simple low-code Agent building tools, Ant Tech's "Super Factory of Agents" has several distinct evolutionary features in its product architecture.
First is standardization and assetization. It no longer allows users to start exploring from a blank prompt box, but instead packages industry's accumulated know-how into job-level expert agent templates and standardized skill sets.
For example, Ant Tech’s Agentar financial version is built upon its successful AI implementation practices in the financial sector, encapsulating industry know-how and validated methods into reusable skills, expert agents, and toolkits.
For financial institutions, this means there’s no need to start from scratch; instead, they can connect their business scenarios, knowledge bases, and data assets to something that is ready to use, thereby reducing duplicated construction costs.
Another concern is fault tolerance for business in sensitive areas.
The greatest challenge in moving agents from "companionship" to "employment" is unpredictable hallucinations, unauthorized actions, and systemic risks. In core scenarios such as financial clearing, credit risk control, or production scheduling, any unauthorized action could be disastrous. Thus, a set of tools must be established to address the risks related to agent digital identity, data and business security, and model robustness.
The focus is no longer on the model itself, but on the completeness of the security and trustworthy infrastructure.
Ant Tech has invested in blockchain, digital identity, privacy computing, and security risk management for a long time; these capabilities are now being integrated into the foundation of the Agent Factory.
Among them, KYA (Know Your Agent) is seen as important identity infrastructure for the age of intelligent agents.
If, in the internet era, KYC and KYB were prerequisites for transactions and collaborations to be established, then in the future age of agents, KYA may also play a similar role. Be it collaboration, transaction between agents, or payment and execution, the prerequisite is to first confirm "who this agent is."
In this sense, KYA is not just an identity management tool but may become a key entry point in the age of agent economics.
During the Bund Conference, Ant Tech, along with over a dozen organizations, launched national standards for intelligent agent identity management projects. Ant Group also simultaneously released trusted identity solutions for agents and KYA products.
In a sense, this is also a crucial foundational infrastructure for the integration of intelligent agents into the core business systems of enterprises, filling in the layers of identity, authority, and responsibility delineation.
Summed up in one sentence: whether agents understand core business logic, whether knowledge can be reused, whether components can be called stably, and whether operational permissions can be controlled determines how much value agents can create within enterprises.
Who can master the Agent Factory?
The answer is hidden within the changes in the industry value chain structure.
As large model technology becomes increasingly standardized, the implementation of AI in enterprises no longer mainly depends on the model layer. The true barriers lie in two things: understanding the scene and being able to deliver it in an engineered manner.
Palantir is the best example of this path. This company does not have its own large models but has become one of the biggest winners in the era of AI.
It relies not on models, but on over two decades of grounding in government and enterprise scenes, transforming clients’ chaotic, non-standard, documentation-poor businesses into a reusable "ontology" and AI productization platform.
When the AI wave surged, it did not need to re-understand clients; it only needed to deliver the accumulated understanding of scenarios in an AI manner, resulting in explosive performance and market value.
Similar traits can be found in Ant Tech.
As an independent arm of Ant Group's technological commercialization, Ant Tech has long served the digital upgrades of industries such as finance, government, transportation, and energy, and has now collaborated with partners to serve over 30,000 enterprise clients across 24 countries and regions, including Asia, the Middle East, Africa, and Europe.
Over the past decade, it has accumulated three hard-to-replicate assets: an understanding of industry scenarios, the product engineering capability to bridge the final mile of technology implementation, and the trust foundation earned through long-term service.
These capabilities and experiences that help enterprises land services have become the barriers for Ant Tech's enterprise AI business, as well as an important starting point for its AI business growth.
Unlike many companies "entering enterprise AI from large models," Ant Tech is coming in from the opposite direction. The latter starts with model capabilities and then looks for scenarios where they can be applied, while Ant Tech is already situated within these industry client scenarios, beginning from the clients' questions and needs.
Ant Tech CEO Zhao Wenbiao stated to reporters that over half of the growth in its AI business comes from existing clients; these major clients have established long-term trust with Ant Tech, which is the most scarce asset today. Data backs this up: the customer net revenue retention rate (NRR) exceeds 110%. The proportion of AI product purchases among major clients has approached 20% and continues to grow.
This also explains why Ant Tech's intelligent agent super factory has first succeeded in financial management, credit risk control, and electricity trading scenarios, which have extremely low fault tolerance yet high value for AI.
For instance, Ant Tech’s "Digital Customer Management Expert" agent can coordinate multiple sub-agents to complete key processes such as customer insight, product matching, strategy generation, and service outreach, automatically generating personalized management plans that compress workflows that previously required cross-role and multi-step collaboration to be completed within a day.
This is not just about efficiency improvement; it has resulted in concrete business effects. In practical applications, the piloted customer group has seen a total asset increase of 10%, customer activity increase of 15%, and a fivefold increase in the number of served customers.
Unlike most AI service providers, Ant Tech possesses specially trained industry large models, a model MaaS platform, and application agent platforms. Most companies in the market focus on one of these areas and rarely integrate these capabilities to directly serve industrial applications. Even Palantir does not have self-developed models.
"In my view, Ant Tech is a very unique company." At the 2026 Bund Conference, Ant Tech CEO Zhao Wenbiao noted that entering the AI era, Ant Tech is not simply producing some products extended to AI, but is focusing on being AI native. Simultaneously, the company's operational model and management indicators are also evolving.
It is reported that over the past two years, the company’s overall business has maintained an average annual growth of 50%, with the AI To B business growth reaching triple digits, becoming a new growth engine. The business model has also gradually transitioned from mostly large client project-based deliveries to platform-driven, standardized product delivery, with platform product ARR growth reaching 60%.
In other words, today, Ant Tech has completed an AI-based transformation.
Third-party reports also confirm this point.
According to IDC's report, by 2025, Ant Tech ranks first in the overall market for large financial models, intelligent agent applications, and services in China; and the "China Intelligent Agent Development Platform Market Share, 2025" report shows that in the private market for intelligent agent development platforms in China in 2025, Ant Tech ranks first among non-cloud vendors and fourth in the overall market.
Reversal of the AI value chain: from "right triangle" to "inverted triangle"
At this year's Bund Conference, a research report released by Wu Yujun, a professor at Shanghai Jiao Tong University and a researcher at Gao Jin Think Tank, pointed out that enterprise-level intelligent agents are bidding farewell to "model supremacy," with the orchestration and governance layer becoming the new growth pole.
Coincidentally, Li Shiwei, head of Ant Tech's International Division, provided a judgment that the value distribution of the AI industry chain is about to flip from "right triangle" to "inverted triangle." This means that as the industry matures, the value proportion of the upstream model layer is decreasing, while the value of the intermediate and AI application layers will rise rapidly.
For AI applications, Li Shiwei further divided them into two layers. The first layer is general scenarios, such as coding and work, where Agent products are very lively and competition is fierce. However, once entering industry scenarios, i.e., production systems, the situation changes completely. In Li Shiwei's view, this is the true deep water zone and main battlefield of AI to B.
Ant Tech's smart agent super factory strategy focuses precisely on this deep water zone.
When model supply is sufficiently abundant and capability gaps continue to narrow, in this new phase, the focus of competition has shifted to who can transform model capabilities into standardized, reusable product capabilities and truly apply them to the core business scenarios of enterprises, continuously creating actual value.
Returning to the initial statement: launching an intelligent agent is easy; the true dividing line for the business of AI realization is no longer about who produces more agents. It is the ability to successfully navigate the complete cycle of "building, managing, using, and governing" that creates end-to-end value that will occupy the next stage’s table.
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。
