At a recent meeting held at Mengniu's headquarters, employees from departments such as sales, R&D, and supply chain took turns presenting how they are using WorkBuddy to transform their daily tasks and boost efficiency.

Beyond the competition itself, Mengniu is now considering nurturing some of these employees toward the role of FDE (Forward Deployed Engineer), aiming to integrate AI agents more deeply into corporate workflows.

This approach is no longer unique to Mengniu. COSCO Shipping Specialized Carriers began implementing an internal FDE model this year, while COSCO Shipping Holdings has also incorporated FDE into its talent development strategy.

On the other side, AI giants like Tencent and Alibaba are dispatching more FDEs into enterprises, hoping these engineers will immerse themselves in business operations, get AI agents running effectively, and bring frontline issues back to refine their products.

A subtle boundary is emerging: AI vendors are using FDEs to push deeper into core enterprise processes, aiming to expose agents to more real-world workflows.

Meanwhile, enterprises are cultivating their own FDEs, seeking to keep judgments about business, processes, and data boundaries more in-house even as they open up their scenarios.

Where this division of labor will ultimately settle remains an open question.

The rise of FDE signals that competition in the office agent space has moved beyond the product itself and into a new phase where tech giants and enterprises are redefining how they organize collaboration.

FDE Takes the Frontline

This year, office agents have become one of the most crowded battlegrounds in the AI race among major tech companies.

Tencent launched the enterprise version of WorkBuddy in June, Alibaba integrated QoderWork, Wukong, and MuleRun into "Qianwen Office" in July and began testing, while ByteDance further consolidated its Feishu and Doubao-related teams.

Though the product forms differ, the direction is converging: AI now directly reads documents, handles spreadsheets, calls software, and then integrates further into corporate knowledge bases, systems, and workflows.

But bringing AI truly into an enterprise is not as simple as installing an agent on a computer.

Beyond IT departments, business staff—though most aware of their work pain points—may not understand the boundaries of model and agent capabilities, nor can they easily judge how a business task should be broken down into AI-executable steps. Conversely, AI companies understand AI best but cannot naturally grasp how orders flow through a given enterprise.

This gap between product and business requires a "translator"—someone who can convert enterprise problems into AI-solvable tasks while also feeding business understanding back into sustainable AI iteration.

This is why FDE has been pushed to the forefront.

FDE stands for Forward Deployed Engineer, typically translated as "前线部署工程师" in Chinese. Compared to traditional pre-sales or implementation roles, FDEs stay closer to the client's real business: they go to the frontline to identify scenarios worth AI transformation, assemble models, agents, data, and systems into functioning workflows, and bring exposed issues back to the product side.

Palantir is one company that has perfected this model, with its FDEs working on-site at client locations to solve specific problems, then distilling recurring business rules, data relationships, and requirements back into the product system.

In this round of office agent competition, this playbook has taken on more direct commercial significance: for an agent to truly enter an enterprise's daily operations, someone still needs to go to the frontline to find scenarios, connect systems, adjust workflows, and drive employees to actually use the agent.

Domestic tech giants have already begun offering hefty salaries for FDEs. Public recruitment postings show that ByteDance's "Doubao AI Large Model FDE" position offers a monthly salary of 35,000 to 70,000 RMB with 15 monthly payments, reaching an annual maximum of around 1.05 million RMB; Ant Digital's B-end FDE role pays 40,000 to 60,000 RMB per month with 15 payments; and Zhipu Huazhang's FDE lead position commands 60,000 to 80,000 RMB monthly.

Behind this high-salary scramble, the FDE's role extends beyond project delivery.

As office agents begin operating browsers, reading documents, and calling enterprise systems, how a task is decomposed, where a tool fails in a step, and why users redo work all form complete feedback loops. For AI companies, genuinely entering these workflows means the product can, for the first time, continuously surface issues in real tasks, driving further optimization of harnesses, models, and products.

Thus, competition around office agents is extending from product capabilities into real workflows. FDEs stand at the closest point to enterprise business, responsible for delivering agents into operations and bringing on-site tasks and issues back.

But as AI companies push further into enterprises, another shift is already underway.

Giants Push Inward, Enterprises Cultivate Their Own

According to recent field visits by Wall Street News' All-Weather Technology team, some companies have begun cultivating their own "FDEs" while introducing office agents.

Mengniu is one such example. On August 11, All-Weather Technology observed 45 employees participating in a WorkBuddy AI application roadshow at Mengniu's headquarters.

In this competition, IT staff were not the primary participants. Mengniu selected 200 "AI Pioneers" from 28 tier-one business units, with only 3 from the Digital Technology Innovation Department; the rest came from sales, R&D, supply chain, and other business functions.

Mengniu deliberately reduced the proportion of IT personnel because business staff "bring their own scenarios." They know best which daily tasks are repetitive, which processes consume the most time, and which steps are most error-prone—yet these individuals may not have known what models and agents can already do, nor possess the ability to redesign workflows with AI.

Mengniu chose to first train business personnel, then identify parts of real work that can be transformed by AI.

According to Mengniu's introduction to All-Weather Technology, the company is considering developing a group of in-house "FDEs" from these business staff and plans to establish an L1-to-L3 AI Pioneer certification system. L3 requires not just AI proficiency but also the ability to understand business needs, build solutions, and track results continuously. Meanwhile, Mengniu's Digital Technology team has already set up a dedicated AI FDE group.

Behind this lies a practical cost consideration.

AI demand within large enterprises is often highly fragmented, with sales, R&D, supply chain, finance, and other departments constantly surfacing new scenarios. If every scenario requires waiting for external engineers to re-understand the business, build solutions, and connect systems, the more scenarios there are, the harder it becomes to amortize delivery costs.

Xu Feixiong, head of Mengniu's low-temperature new retail business, acknowledged that business teams already carry performance pressure and cannot afford to add a separate batch of IT staff just for AI efficiency gains. More often, original business personnel need to learn how to use and adapt AI themselves.

But the deeper significance may lie in enterprises wanting to embed this capability within their own systems. As agents begin touching core processes like orders, supply chain, R&D, and finance, FDEs get closer to a company's data and business assets—capabilities that are difficult to leave entirely in the hands of external teams over the long term.

Of course, this security and data boundary is also on the radar of the tech giants.

Min Liming, a Tencent smart retail industry solutions expert, told All-Weather Technology that the earlier phase focused on getting employees to use WorkBuddy for personal repetitive tasks; the next step will enter the "deep water zone."

"The next stage enters the deep water zone, and frankly, it's a particularly challenging one," Min said. Tencent will work with Mengniu's Digital Technology team to jointly assess which systems and data can be opened to AI, which systems hold the highest data value and should be prioritized for integration, and gradually connect them via MCP. At the same time, model calls must pass through Mengniu's internal large model security audit gateway, with data connections established strictly within security boundaries.

Mengniu is not an isolated case.

COSCO Shipping Specialized Carriers began implementing an internal FDE model in March this year, directly assigning 18 digital professionals to stay at business frontlines, participate in business meetings, visit workstations, and map out specific processes such as bill of lading processing, voyage scheduling, container management, and semi-submersible vessel stowage, breaking down the barriers between "technology" and "business." COSCO Shipping Holdings, meanwhile, has already written FDE into its 2026 corporate talent development direction.

These changes sketch a more nuanced relationship in enterprise AI deployment: tech giants use FDEs to move deeper into enterprises, while enterprises hope to retain the most critical business understanding in their own hands.

Co-Building: Perhaps the More Practical Solution

Seen this way, FDE does not necessarily have to be an either-or choice between "external deployment by giants" and "internal building by enterprises."

As agents enter deeper business processes, a more realistic collaborative model may be co-building FDE capabilities.

AI vendors possess engineering capabilities like agent harnesses but need real workflows to continuously expose issues; enterprises understand their own business processes, data permissions, and organizational rules best, but also need external technology to get these scenarios running.

Neither side alone can easily complete the full chain from scenario discovery, system integration, to continuous iteration.

For AI vendors, this co-building model has another layer of practical significance.

Office agents ultimately need to be a product business, not a labor-intensive business that expands by adding more on-site engineers.

The more important value of FDEs entering enterprises is to first identify problems that standard products cannot yet cover, then gradually distill those problems into harnesses, skills, connectors, and product capabilities.

Ideally, the same type of problem might require repeated FDE debugging at the first client, but by the second or third client, more parts should be directly reusable.

Only then can frontline delivery continuously feed back into the product, and the marginal returns of office agents become "thicker" as customer numbers grow.

From this perspective, co-building FDE may first suit large enterprises like Mengniu. Large enterprises have more complex systems, longer processes, and more real scenarios, and are better positioned to allocate their own business, data, and technical personnel for long-term collaboration with AI vendors.

Large enterprises can also serve as a "testing ground" for the continued productization of office agents. Within legal compliance and clear data boundaries, the specific problems FDEs solve with enterprises can continue to be distilled into harnesses, skills, and product capabilities.

What agents thus learn is not the "family secrets" of one company, but methods for handling a class of enterprise problems.

If the workflows that today require FDEs to run through repeatedly with one large enterprise can increasingly be handled directly by the product in the future, then small and medium enterprises will not need to replicate the same heavy delivery model.

In some sense, co-building FDE may be the more realistic middle ground in this delicate game: AI vendors keep advancing, and enterprises do not have to hand themselves over entirely.