A group of seasoned technical engineers, deeply rooted in China's software industry, are attempting to rebuild enterprise-grade software using the latest technologies and methodologies. This is the entrepreneurial story of Deepexi Technology, a leading domestic enterprise in large models.

As the helm of the company, Zhao Jiehui, founder, chairman, executive director, and CEO of Deepexi Technology, has developed new insights after witnessing dramatic changes in the industry firsthand in recent years.

When discussing transformations in the software industry, Zhao Jiehui still points to Palantir first. In his view, the company's core value lies in breaking a fundamental misconception in traditional software: enterprises don't buy software for attractive interfaces or standardized processes but to effectively link scattered knowledge and data.

Over the past two to three decades, traditional software companies have focused almost entirely on "interface creation." Yet, the real pain points of enterprises remain unaddressed: data sits idle in systems, knowledge is fragmented across departments, isolated and unable to form synergies, ultimately becoming useless numbers and documents. Palantir's approach is straightforward—relying on Forward Deployed Engineers (FDEs) to manually construct knowledge logic networks for enterprises, connecting dispersed business nodes, data, and expertise.

However, this model has its flaws: heavy reliance on manual labor leads to high costs, long cycles, and poor iteration capabilities.

Against this backdrop, a new idea emerged within Deepexi Technology: Why not take the next step forward, standing on the shoulders of giants?

01

From Knowledge to Skills

Ontology Modeling: The Key to Digital Employee Implementation

In Zhao Jiehui's logic, the transition of software companies to AI companies isn't about creating simple AI applications but achieving an evolution from "Knowledge" to "Skills." Behind this lies three generations of shifts in enterprise IT construction logic.

Enterprise IT construction has undergone three core stages.

The first was represented by SAP + Oracle databases, focusing on indicator modeling + business software. It relied on data platforms at the base and software interfaces for unstructured data governance, encapsulating business flows, approval logic, and decision-making logic into fixed pages. The second stage, introduced by Palantir, incorporated data lakes and ontology modeling, transforming data governance from large, flat-table structures into semantic networks with objects, relationships, rules, and execution. However, this process was entirely manual, carried out by FDE engineers.

Deepexi Technology is now advancing the third stage: fully automating ontology modeling—previously done manually by FDEs—through models.

"Many so-called enterprise AI solutions today remain stuck at the knowledge management stage, like a college graduate who has learned mechanical knowledge but can't apply it—because their mind lacks the industry- and enterprise-specific knowledge links, which we call ontology modeling," Zhao Jiehui explained.

This is the core of Deepexi Technology's product launch, offering a deeper upgrade to Palantir's ontology modeling.

Drawing on experience serving hundreds of top-tier clients, Deepexi Technology has solidified knowledge logic frameworks and semantic networks for key industries like manufacturing and consumer retail. Through specialized training on industry causality datasets, its models can now automatically perform ontology modeling.

For automotive companies, for example, engineers no longer need to spend months building models. By inputting raw data like drawings, process documents, and supply chains, the model automatically generates a dedicated ontology model. This model isn't a table or chart but an intelligent brain capable of "acquiring skills."

Zhao Jiehui explained that based on this ontology model, capabilities like on-site image recognition, supply chain management, and order replenishment analysis become specific skill modules (Skills). Combined, these modules form AI digital employees tailored to different enterprise roles.

"This is the true transformation from knowledge to capability. Models without ontology modeling can only store and retrieve knowledge. Only those with ontological awareness can truly work for enterprises, becoming deployable AI digital employees," he said.

He drew an analogy: Professional knowledge learned in college is foundational, but the ability to apply it to solve real-world problems in an enterprise is what matters. Enterprise ontology modeling bridges the two. Deepexi Technology's goal is to empower models with this bridge, turning enterprise knowledge into problem-solving capabilities.

Supporting this closed loop (closed loop) are two core capabilities of Deepexi Technology's enterprise large model: AI for ontology-based data governance and AI for coding.

The former accurately parses enterprise multimodal, deeply nested complex data, establishing cross-document, cross-department knowledge semantic networks. The latter rapidly generates executable skill modules based on ontology models, enabling digital employees with flexible business execution capabilities.

02

Model as Software

Enterprise Large Models as the Core Delivery

The software industry in 2026 is being propelled into a new era by two forces.

On one side, Palantir redefines enterprise data value chains through ontology modeling. On the other, OpenClaw disrupts traditional software IO bottlenecks with lightweight intelligent retrieval. This revolution, driven by foundational technological shifts, is rendering the three-decade-old "interface + database" model obsolete.

"OpenClaw's greatest success is making the industry realize that data doesn't need to be neatly organized into databases—unstructured raw data can also create value," Zhao Jiehui said. This aligns perfectly with the urgent needs of most enterprises: not all data requires massive-scale databases. Department- or team-level multimodal data often benefits more from lightweight solutions.

Traditionally, software assumed "data must be structured for storage." To meet this, enterprises expended significant manpower inputting PDFs, Excel sheets, and handwritten notes into systems—a tedious, inefficient process that lost valuable information. Meanwhile, as data volumes grew, manual organization became impossible, forcing reliance on software and interfaces for standardization and storage.

Now, OpenClaw has overturned this assumption: models' unstructured data processing capabilities have become strong enough. Excel tables, PDF documents, Word summaries, even handwritten notes and on-site videos, can now serve as direct data inputs, eliminating the need for laborious database entry.

However, OpenClaw only addresses individual-level data governance needs. Enterprise scenarios are far more complex.

Enterprise data differs from consumer data in two key ways: First, it is highly formatted and deeply nested, including complex tables, formulas, drawings, and 3D spatial data. Second, strong interconnections are essential. For example, a formula in a drawing might be explained in another document, with parameters tied to a complete knowledge system—hence the need for ontology modeling.

This technological paradigm shift is at the heart of Deepexi Technology's product system upgrade.

Built on its self-developed FastData Foil enterprise data fusion platform, Deepexi Technology can parse and govern all enterprise data types. Its enterprise large model, Deepexi 2.0, automatically generates ontology models, which the FastAGI platform then transforms into deployable skill modules and digital employees.

Zhao Jiehui cited a practical example: "Previously, enterprises analyzing supply chains had to manually input purchase orders, acceptance forms, and logistics records into ERP systems to generate reports. Now, they simply store these files in a data lake. FastData Foil parses all data, and the model analyzes it based on ontology logic, delivering results in minutes— dozens of times (dozens of times) more efficient than traditional ERPs."

More importantly, this approach completely restructures software delivery.

Under this logic, what customers receive is essentially an enterprise large model paired with supporting tool components—no complex front-end pages or cumbersome systems. Like a raw apartment, Deepexi Technology builds the framework, and clients "renovate" it by inputting their specific data, creating a tailored model. This is the essence of the AI-era software revolution: shifting from selling tools to selling capabilities.

"Pre-compiled enterprise business software will gradually become obsolete," Zhao Jiehui predicted. "In the future, the software industry won't have distinct categories like ERP, CRM, or OA. All software forms will converge into models. When enterprises buy software, they're essentially buying model capabilities—the intelligence to understand them, solve problems, and continuously evolve."

03

Becoming the "Windows" for Enterprises to "Raise" AI Employees in the AI Era

At its core, regardless of future changes, enterprises always prioritize cost, security, and implementation effectiveness.

With Deepexi Technology making enterprise models the core delivery form, these three concerns now have innovative solutions.

Regarding security, the permission management logic of enterprise models is fundamentally similar to traditional enterprise security—only more flexible. Models can be deployed locally, with data stored on the enterprise's own servers. Deepexi Technology has also collaborated with Southern University of Science and Technology to train dedicated security models, ensuring robust permission management and data privacy.

On cost and business models, Zhao Jiehui was clear: "The future enterprise service market will shift from 'man-day pricing' to 'token-based pricing.' Man-day pricing charges for processes regardless of outcomes, whereas token-based pricing charges based on actual value created—how many problems the model solves and how much value it generates. This is fairer for enterprises."

Token consumption is inherently tied to enterprise development. Zhao Jiehui explained: "If an enterprise merely maintains the status quo without new knowledge breakthroughs or business logic, token consumption remains fixed, with annual costs staying level. But if the enterprise innovates and explores new knowledge boundaries, training new logic and frameworks into the model will generate additional token consumption—essentially paying for innovation."

He emphasized that Deepexi Technology's core philosophy in enterprise-grade model services is to liberate humans. Let models handle execution and repetitive tasks, freeing people to think, innovate, and explore new knowledge frontiers. Humans then refine these new insights and train them into the model, which operates based on the new frameworks—a virtuous cycle.

Built on this comprehensive product system and business logic, Deepexi Technology has refined its product lineup into DeepexiOS, an AI-grade enterprise operating system.

"All enterprises today want to leverage large models, with the core goal of deploying digital employees. Future enterprise IT construction will essentially involve configuring digital employees for various roles. You need an infrastructure like DeepexiOS to manage the full lifecycle of AI digital employees," Zhao Jiehui said.

This operating system comprises three core components: Deepexi Technology's enterprise large model, Deepexi 2.0; FastData Foil; and FastAGI. Enterprises simply input business data for corresponding roles and domains, and the system automatically performs ontology modeling, generates skill modules, and assembles them into deployable digital employees.

"In the previous era, enterprises needed various software to manage resources and processes. In the next era, they'll need AI-grade operating systems like DeepexiOS to generate and manage digital employees," he said. In his vision, this system will evolve into an enterprise-grade AI operating system—not just a tool but core infrastructure for enterprise AI.

"Many now debate how to make new AI technologies compatible with old software systems. This is like replacing green trains with high-speed rail—not because green trains are bad, but because high-speed rail's underlying logic is superior. Software industry transformation follows the same principle: it's not that traditional software is inferior, but that models better align with AI-era enterprise needs—liberating humans for innovation while machines handle execution," he said. Hence, "model as software."

As enterprise large models grow more powerful, demand for traditional Agent tools will decline. "The core of enterprise AI is enhancing the model's understanding of the enterprise, not building compensatory tools around it. You can call a combination of skill modules an 'Agent,' but it's the model's inherent capability, not an external overlay. This differs from current Agent tools in the market."

The entire large model industry is also clearly differentiating. "The mystique around large models has faded. Training methods and engineering logic are no longer barriers, nor is building ultra-large models the only path. Eventually, each large model will find its niche—C-side models may focus on social or coding, while enterprise-side models will specialize in deploying digital employees. By year-end, you'll see that leading model companies won't just offer models but operating systems for their respective domains."

Palantir and OpenClaw are merely catalysts for the revolution in large-scale enterprise AI adoption. The real shift is that enterprise needs have evolved from "processing data with tools" to "solving problems with capabilities." As Deepexi Technology's founding mission states: in enterprise services, there's no need to chase temporary industry consensus. True innovation leads, creating new consensus through tangible results.