Zoho Releases Enterprise-Grade AI Full-Stack Three-Layer Architecture: AskZia, Zia Agents, and Agent Studio

On October 20, at Zoho's Beijing media briefing themed "Empowering the Global Growth of Chinese Enterprises," Zoho not only showcased its latest enterprise-grade AI panoramic strategy to the outside world, but also sought to answer a common pain point in the industry today: after two years of hype around AI large models, why do most enterprises still feel that AI "can't be put to use" in their business?

Zoho Global Vice President Mathivanan pinpointed the crux of the problem: "Too short a decision cycle, too many candidate tools, and from the very beginning, too many data silos." When enterprises blindly stack various software products bearing AI labels, not only do they fail to break down the original information silos, but they instead give rise to "AI silos."

Zoho China COO Xia Haifeng was blunt in his subsequent keynote: "If AI can only answer questions and provide information, but cannot participate in actual business processes, its value to business is minimal."

This also reveals a key threshold facing enterprise AI applications: being able to answer questions does not equal being able to complete tasks. For AI to truly participate in business, it not only needs to understand users' needs, but also needs to obtain relevant data, call business systems, and complete specific operations under the constraints of permissions and processes.

To cross this gap from "understanding how to answer" to "being able to execute," Zoho systematically unveiled at this event its three-layer AI panoramic architecture accumulated over many years: data at the bottom layer, connectivity at the middle layer, and execution management at the top layer.

Crossing the Watershed: From an Assistant That "Can Talk" to a Digital Employee That "Can Do"

In Zoho's AI panorama, the top layer is the core that directly reaches the business. Zoho believes that the true AI watershed lies in whether it can take on tasks and run through multi-step processes.

First is AskZia, a conversational AI assistant natively embedded in business scenarios. It is not an independent bolt-on tool, but is deeply rooted in systems such as CRM, finance, and customer service. Users can obtain information and analyze business data through natural language, and complete work such as content drafting and polishing with the support of corresponding functions.

But "being able to talk" is only the first step; "being able to do" is the real rigid demand for enterprises. Zoho launched preset agents Zia Agents, with hundreds of out-of-the-box roles such as sales coaches and SDR bots, specifically handling work that is regular, reusable, and experience-intensive.

For scenarios with extremely complex rules and strong personalization, Zoho provides the Agent Studio low-code platform. For example, an auto parts customer used this platform to develop its own dedicated cost forecasting agent, effectively responding to material cost fluctuations caused by factors such as tariffs and transportation in international trade. This ability to let non-technical personnel customize exclusive AI like building blocks is precisely Zoho's differentiated moat compared with standardized SaaS products.

From Ask Zia to Zia Agents, and then to Agent Studio, the three respectively correspond to information acquisition and processing, specific task execution, and agent construction for personalized processes. Their difference lies not only in the strength of AI capabilities, but also in the change in the way AI participates in business: from assisting employees in completing work, gradually extending to undertaking specific tasks.

Connectivity and the Bottom Layer: Not Chasing Parameters, Only Building "Native Intelligence" Suited to Business

If the AI execution layer is the part of the iceberg above the water, then data and connectivity are the massive foundation below the water. At the middle layer, Zoho has opened up internal and external ecosystems through MCP (Model Context Protocol).

At the event, Xia Haifeng depicted such a scenario: users configure custom connectors in MCP-supporting clients such as WorkBuddy, Doubao Work, and Qianwen Office, connect to Zoho MCP, and can directly operate Zoho CRM through conversation externally, achieving automatic data updates. Zoho MCP is currently compatible with mainstream large models and agents such as Claude and ChatGPT, and provides an integration list of more than 300 third-party services. Through standardized protocols, Zoho not only enables its own business capabilities to be called by various external agents, but also ensures that sensitive data is masked and authenticated before leaving the system, achieving true openness and security in parallel.

In addition to system connectivity, the models themselves also need to adapt to different business tasks.

Today, when computing power and models routinely pursue tens of billions or even trillions of parameters, Zoho has chosen an extremely pragmatic path at the bottom layer. Mathivanan said that Zoho's investment does not follow competitors' money-burning scale, but is determined by customers' real needs. Compared with general large models that consume enormous resources, Zoho prefers self-developed small and medium-sized models deployed in its own data centers. "For example, in scenarios such as automatic email replies, image content summarization, or sentiment analysis, it is enough to deploy suitable models. This not only improves business accuracy, but also greatly reduces inference costs."

This ultimate control over business boundaries and costs is also reflected in the application of AI programming. Faced with the question of whether AI writing code will replace traditional software, Xia Haifeng answered with the "iceberg theory": AI programming is only the tip of the iceberg above the water; the invisible parts such as source code hosting, permission design, security operations, and underlying data governance are where the real effort lies. Zoho empowers users with AI programming capabilities so that they can extend on the existing, solid business foundation, rather than building an application suspended in midair out of thin air.

In terms of business model, Zoho's choice is likewise "counterintuitive." When some vendors try to charge customers high fees for AI capabilities as an independent package, Zoho argued at the event that AI should be integrated into the product system rather than simply regarded as an independent value-added feature, and that it currently does not charge customers extra for AI.

Behind this decision is the long-termism confidence of Zoho's 30 years of adhering to no financing and no IPO. For enterprise-grade AI, the significance of underlying investment lies not only in providing computing power, but also in supporting product operation, data management, and collaboration among different business applications.

Data security is another foundational requirement that cannot be avoided. Mathivanan defined it as "the bottom line of the transition period": whether using an enterprise's own API Key to access external models or using Zoho's self-developed models, Zoho promises not to use customer data to train general models. All technological exploration ultimately points to one goal—protecting customers' data security and privacy.

Therefore, enterprises deploying AI should not focus only on model capabilities, but also need to evaluate data processing methods, permission configurations, and the boundaries of access to external services in light of actual application scenarios. For software vendors, the degree of openness of AI capabilities and data protection mechanisms are not two independent matters: only by clarifying access scope and execution permissions can enterprises be better positioned to use AI for real business.

As Xia Haifeng concluded at the end of his speech: "As agents gain the ability to call APIs and autonomously execute tasks, enterprise software is no longer just a functional tool for people to operate, but also becomes the foundation for carrying agents to conduct business. For enterprises adopting AI, it is not about choosing a smarter model, but about deciding which links to hand over to machines and what kind of foundation to prepare for that. What Zoho wants to do is truly implement 'AI + autonomous execution' across all enterprise scenarios, rather than tell stories."

AI is changing the way enterprise software is used and is also pushing SaaS to rethink its own value. In the future, competition in SaaS will lie not only in how many features are provided, but also in whether AI can truly participate in business and translate technological capabilities into actual business value.

Leiphone