Is it reliable for business leaders to use AI to safeguard their wealth?
This summer, the experience of a private software company in Zhangjiang, Shanghai, sparked significant waves in the AI industry.
Rumors spread online that in pursuit of "talent internationalization", this company recruited an Indian R&D team lead and granted them the highest-level access to the code repository. In just four months, this lead successively brought seven of their former classmates into the team.
The core development team of a Chinese company turned into a "hometown association" for Indian programmers. Internal communications in some scenarios even switched to Hindi, completely sidelining Chinese employees and planting hidden risks.
When the company failed to meet their nearly doubled salary increase demand, the fuse of conflict was ignited. The lead directly locked all core code repositories, revoked all access permissions of more than 40 Chinese programmers, modified global keys, and appropriated over a year of programming achievements for themselves, bringing the project to a 48-hour halt with imminent risk of breach of contract compensation.
Eventually, it was reported that the Chinese operation and maintenance team bypassed the permission layer, worked overnight to reset keys and regained control, and the eight involved employees were dismissed. This incident exposed a shocking problem: a company's core digital assets could be so vulnerable.
In the AI era, what business owners call "wealth preservation" does not refer to money, but more to "digital family property". However, the core secrets and lifelines of many enterprises today, just like this Zhangjiang company, are hanging by a thread due to the emphasis on development while neglecting operation and maintenance.
This incident further accelerated the process of "AI entering companies to help business owners preserve their wealth". Major AI giants have also set their sights on this lucrative market segment.
The Dilemma of Wealth Preservation
In 2006, 35-year-old Zhou Hongyi launched 360 Security Guard. At that time, malicious software was rampant across China's internet. 360 disrupted the industry with its "free security" strategy, helping hundreds of millions of enterprises and internet users automatically detect and remove malicious plugins and fix security vulnerabilities on their devices.
But 20 years later, the traditional "free antivirus + firewall" model can no longer guarantee enterprise security. The default "internal trusted" logic of firewalls has long been broken, replaced by massive loopholes arising from "internal enterprise permission management".
This Zhangjiang enterprise incident is not an isolated case.
In August 2025, the "Zunpai case of stealing Huawei's chip trade secrets" exposed by the media was a similar example. After Zunpai, a company engaged in so-called high-end chip design, was established in 2021, it was revealed to have systematically poached entire teams of Huawei employees, even leading to situations where "some people worked at Huawei during the day and went to Zunpai at night".
In addition, there was the "19.9 yuan baby bottle case" in Ningbo, Zhejiang in 2024, where a departing employee changed the price of baby bottles originally priced at over 100 yuan to a "dirt-cheap price" of 19.9 yuan, causing losses to the company. In 2018, a former network administrator of a children's products company in Shanghai, out of dissatisfaction with the company, maliciously changed all product prices on the e-commerce platform to 10% of their original value, resulting in losses of over 2 million yuan.
These cases reveal new characteristics of enterprise risks:
Traditional security measures — firewalls, access cards, confidentiality agreements — are designed to guard against "outsiders". Today, a large number of risks come from within, stemming from complex human nature and subtle interest entanglements.
Some tech enterprises that should focus on internal management often prioritize R&D over protection, easily granting the highest permissions to a tiny number of people, and lacking hierarchical control, permission revocation, and dual backup mechanisms.
This exposes truly "priceless" digital assets such as source code, core algorithms, customer data, and supply chain information to threats, making losses highly likely to occur.
In 2025, the People's Court of Bao'an District, Shenzhen released the "Judicial Protection Report on Trade Secrets (2020-2024)", showing that over 80% of the trade secret cases it heard were related to employee departures, with disputes concentrated in core positions such as sales and technology.
This fully demonstrates that CTOs or core technical personnel leaving for new jobs with confidential information has become the most common scenario of information leakage. Once personnel change, enterprise assets are easily lost.
As firewalls, access cards, and confidentiality agreements appear powerless in the face of human nature's flaws, more and more enterprises are turning their attention to a new "gatekeeper": AI.
The AI Gatekeeper
In 2023, a company named Tomoro was founded in London, UK, positioned as an AI application consulting and engineering service provider. Its core business is not developing foundational large models, but helping large enterprises truly embed AI into their business systems.
In short, many enterprises put their purchased AI tools on the shelf after procurement, failing to truly match their own needs and generate practical value. Tomoro's value lies precisely in bridging the "last mile" of AI application implementation in enterprises.
Its killer advantage is its team of 150 "Forward-Deployed Engineers (FDEs)". Instead of sitting in offices to fine-tune general models, they are directly dispatched to client enterprises to review business process documents, interview frontline employees, extract data from Excel spreadsheets and orders in ERP systems, and then leverage AI model capabilities to rebuild a functional, customized workflow.
This model can be called "the FDE model — planting engineers on-site at the client".
In May 2026, optimistic about the prospects of this business, OpenAI announced the establishment of a new entity named "The Deployment Company" with initial investment exceeding 4 billion USD, and planned to acquire Tomoro.
OpenAI's calculation is that enterprise AI budgets will not stay at API interface calls for long. The real "large budgets" are hidden in process transformation, system integration, organizational collaboration, and long-term operation and maintenance. This system can not only help enterprises improve internal efficiency, but also clarify ownership of business processes and help business owners protect their assets.
According to estimates from business data platform Prospeo, Tomoro's annual revenue is approximately 8.469 million USD (equivalent to over 60 million RMB), with per capita revenue of about 86,000 USD. However, these figures are not officially disclosed. Multiple media reports have consistently confirmed that Tomoro achieved more than tenfold monthly revenue growth in the 12 months before being acquired by OpenAI, fully demonstrating the sector's booming market.
Figure | What OpenAI values is Tomoro's implementation capabilities
OpenAI's long-time rival Anthropic did not lag behind. In May of this year, it partnered with Blackstone, Goldman Sachs, and Hellman & Friedman to jointly establish a new enterprise AI service company, dispatching engineers to enterprises for customized deployment, which has just started operations.
In China, there are also many companies aiming to become the "Chinese version of Tomoro", each with their own characteristics.
First type: Local practice of the FDE model in China
If Tomoro represents the path where engineers go deep into enterprise sites to connect AI to real workflows, then ForFlow, a company founded in Shanghai, is conducting similar explorations.
It does not simply sell models or tools. Instead, through its self-developed "evaluation large model", it first conducts a comprehensive assessment of the enterprise's business processes, data foundations, permission boundaries, and input-output ratios, then around three links: diagnosis, deployment, and growth, assists enterprises in identifying truly implementable AI investment scenarios.
At present, ForFlow's FDE team has entered on-site scenarios in industries such as electric power, e-commerce, energy, and healthcare, helping enterprises clarify "which scenarios are worth transforming with AI", "how to calculate returns", "how to set up data permissions, approval boundaries, manual takeover mechanisms, and responsibility tracing systems", and precipitating these capabilities into reusable, scalable exclusive AI assets.
Chen Zezhou, founder of the company, stated: "What the industry is solving now is not whether enterprises have AI, but whether AI can stably enter workflows, produce measurable results, and leave behind reusable capabilities."
According to project reviews provided by ForFlow, in an implemented scenario at a certain electric power enterprise, after transforming the relevant business flow, energy consumption costs decreased by approximately 15%-30% compared to the original baseline, and the comprehensive energy efficiency of the auxiliary systems improved by about 10%-40%.
Second type: The Multi-Agent Paradigm
01.AI, founded in Beijing by Li Kaifu in 2023, chose another lighter-weight path.
In January of this year, 01.AI upgraded its "Wanzhi Enterprise Large Model One-Stop Platform" to version 2.5, with the core selling point being "Multi-Agent".
Enterprises only need to input one instruction, and an AI virtual team consisting of visual designers, marketing managers, content managers, media specialists, etc., can be generated, who synchronize their professional knowledge and process progress in real time.
The FDE model sends engineers on-site to break down business processes manually. In contrast, 01.AI's Multi-Agent paradigm embeds business decomposition capabilities into the AI system, allowing AI itself to act as a director scheduling subordinates.
01.AI refers to this transformation as a leap from "talent dependence" to "capability softwareization". When agents decompose, reconstruct, and encapsulate top talent's capabilities into reusable capability modules, enterprises are no longer trapped in the talent cycle of recruitment, training, and turnover.
Third type: Self-Evolving Agents — Using Algorithms to Find the "Global Optimal Solution"
Baidu launched "Baidu Famou" in 2025, whose core capability is not to "execute" tasks, but to "continuously find better solutions".
For example, in the operation of automated port terminals, the problems encountered are: how to arrange containers to prevent cargo ships from tipping over? How to schedule operations to minimize congestion rates?
Baidu Famou's operation neither relies on on-site manual decomposition nor AI autonomous team formation, but focuses on "global optimization under constraint conditions".
Business leaders, known as "veteran experts", can use natural language conversations to guide Baidu Famou to continuously find the optimal business solution. All the business logic accumulated during their interactions with the system, such as production line maintenance dates and special priority for VIP orders, is precipitated as reusable AI assets. Even if the "veteran experts" leave or retire, their tacit knowledge remains in the system.
From Cost to Profit: AI's "Turnaround Battle"
"Business owners are not unwilling to use AI, but they are reluctant to invest in AI projects with unclear returns, uncontrollable risks, and no retained capabilities."
Chen Zezhou of ForFlow believes that many people misunderstand business owners' real demands for AI. What enterprises truly care about is no longer "whether they have accessed AI" or simply replacing manual labor, but "which scenarios are worth doing, which data cannot leave the company, how to accept results, and how to retain experience within the enterprise".
In his view, if AI projects only stay at the stage of a one-time pilot or a single tool, it will be difficult for them to enter enterprises' core workflows. Only when business scenarios, permission boundaries, responsibility tracing, and re-evaluation mechanisms are designed together can AI transform from an external capability into the enterprise's own asset.
He has connected with many enterprises on the front line and found that due to the high cost-effectiveness of Chinese labor and many AI tools getting stuck at the "last mile of business operation", the actual manual replacement caused by AI is still limited.
"For business owners, preventing enterprise risks takes priority over cost reduction. At the same time, using AI to reshape business processes, many AI companies are striving to transform themselves from cost centers into profit centers. This conceptual shift is more welcomed by business owners." Chen Zezhou stated.
Behind this conceptual evolution is the exploration and transformation of AI monetization models.
Before 2024, enterprises rushed to access general large models such as ChatGPT and Wenxin Yiyan to improve employees' efficiency in making PPTs and writing copy. However, although these AI tools helped employees "work faster", they did not help companies "earn more".
Between 2024 and 2025, we entered the monetization phase of vertical models and API calls. AI companies began training vertical models for specific industries, and enterprises paid based on Token usage.
But Li Kaifu pointed out in media exchanges that the Token model works well for scenarios reaching 80-85 points, but for enterprises' "particularly difficult, high-value tasks", clients expect AI to reach 99 points, which AI cannot achieve. Tokens become a cost rather than a benefit.
Now, after the second half of 2025, AI companies have finally realized that the real big money does not lie in selling Tokens, but in helping enterprises transform their processes, integrate systems, and optimize decision-making.
Under this consensus, a large number of AI products embedded in workflows are emerging both domestically and internationally.
In China, there is Xiaoe-tech following the "embedded" path, which directly implants AI Agents into private domain operation systems, covering the entire customer journey of lead generation, conversion, operation, and transactions; there is also MofuXun following the "platform" path, building a general agent middle platform, with AI-related revenue exceeding 100 million RMB in the first half of 2025; there are also Huawei Cloud and ByteDance following the "foundation" path, focusing on underlying computing power and large model support. There are AI operating systems like Feishu, which does not replace old systems with a new one, but builds an underlying operating system for all of the enterprise's "information silos" and "experience silos", enabling them to connect, flow, and eventually be invoked by AI.
These explorations are bringing wealth imagination to the relevant companies.
Li Kaifu disclosed in 01.AI's third anniversary all-staff letter: In 2025, the company achieved 500 million RMB in orders and 250 million RMB in audited revenue. In the first five months of 2026, orders exceeded 1.5 billion RMB, with the goal of going public in 2027.
Companies whose business flows have been reshaped by AI have also gained "considerable profits". YTO Express embedded hundreds of agents into its entire business processes including operations, network management, and customer service. The "Intelligent Routing" agent alone saved the company 152 million RMB in operating costs in 2025; the total quantifiable cost savings created by hundreds of agents in the company reached the "billion-level" in one year.
A state-owned enterprise consulting company structured over 160 scattered project fields through Feishu Projects, connected approval and expense control systems, enabling the entire "project-budget-reimbursement" process to flow automatically, cutting the project initiation cycle by 50%.
"The value of AI is not to run successfully once in a demonstration, but to run stably in the enterprise's daily workflows," Chen Zezhou said. "In the future, truly valuable AI companies will not just deliver a tool, but help enterprises reorganize their demands, data, processes, and people."
From "wealth preservation" to "wealth creation", the story of AI being embedded into workflows may have only just begun. That incident in Zhangjiang was just the opening footnote of this story.
This article is from the WeChat public account "Muhe's Tech Launch Event", author: Gong Zheng, editor: Yang Jing, published with authorization from 36Kr.