At the 2026 World Robot Conference, when asked about the suddenly popular FDE (Forward Deployed Engineer), MingLue Technology CEO Wu Minghui first turned the clock back more than a decade.
"We were already studying this very seriously 12 years ago." When Wallstreetcn's All-Weather Technology asked him about the difference between FDEs and traditional software deployment, Wu Minghui said both involve going to customer sites, but today's FDEs need to go deeper: connecting AI agents into real business operations on one hand, while feeding the capabilities formed in the field back into the backend on the other.
Behind this is a new shift in how enterprise services are delivered.
In May this year, OpenAI established a dedicated Deployment Company, sending FDEs into enterprises to work with customers on requirements discovery, system design, development, and launch;
In June, Anthropic announced a partnership with IT services provider DXC, which plans to train tens of thousands of Claude-certified FDEs to connect Claude into core systems in banking, aviation, insurance, manufacturing, and other industries.
On August 18, Tencent Cloud also directly renamed its original "AI Application Engineer Certification for Agent Development Platform" to "ADP Forward Deployed Engineer (FDE) Certification."
Behind these companies' focus on FDEs lies the "last mile" problem of enterprise AI adoption.
Big players like OpenAI and Tencent have begun personally taking on more enterprise delivery work, dispatching FDEs to connect models and agent platforms into customer systems; enterprise service providers like MingLue Technology are using AI to redo their original custom development, turning industry know-how, data governance, and software implementation capabilities into individual AI agents; and enterprises, as the paying party, have also begun cultivating their own AI pioneers from business departments like sales, R&D, and supply chain, building internal FDE capabilities.
AI has made writing code increasingly cheap, but has turned "who understands the enterprise, defines requirements, and gets agents actually running" into a new business.
One
No Standard Answer
FDEs existed long before the era of large language models.
Palantir early on had engineers go deep into customer sites in a similar way, combining complex data, software capabilities, and specific business needs. The software industry has also always had implementation consultants, solutions architects, and on-site engineers. So after FDEs became popular again in this wave of AI, a controversy has persisted: does it represent a new engineering model, or does it simply follow the old logic of the previous generation of software implementation?
"We already have a pre-sales technical support team. I think FDE is just a trendy term. But if our team believes using this title can help sell products better, I think that's fine too." The CEO of a cloud vendor in South China once told All-Weather Technology bluntly.
Wu Minghui, however, believes that traditional implementation teams can simply be responsible for delivering a fixed piece of software, with no direct connection to the software R&D team; FDEs, by contrast, need to keep contributing capabilities formed on the front lines back to the backend, turning problems exposed at one customer site into assets that can be reused the next time.
This is also why big players like OpenAI and Tencent have begun sending engineers directly to customers. For example, OpenAI's job definition for FDEs clearly goes beyond the traditional "technical support" boundary. Its FDEs are responsible all the way from requirements discovery, technical scoping, and system design to production launch, and must also distill proven methods into tools, playbooks, and building blocks, then bring field feedback back to the product and model teams.
By exposing problems through real workflows, AI giants can continue to optimize their harnesses, models, and products accordingly. But whether this can solve the "last mile" problem of enterprise AI adoption remains an open question.
Wu Minghui gave an example from the advertising industry: even when both serve the fast-moving consumer goods sector, the processes and tactics of P&G and Unilever keep changing.
In his view, expecting a general-purpose foundation model to adapt to all scenarios, handle all planning well, and decompose workflows properly is a paradox. Enterprises must constantly create new processes and tactics to compete; the ways of working learned today may have changed by tomorrow.
Against this backdrop, FDEs face constantly changing context, making it hard for any static "industry answer" to provide long-term coverage.
This also leaves room for another type of company—software service providers.
Two
The Leverage Ratio Test for FDEs
Since AI moved from chatting to "getting work done," the market has harbored concerns about traditional SaaS being replaced. A popular judgment in tech circles is that as vibe coding continues to lower the barrier to development, enterprises will in the future be able to write their own code and build their own agents, further squeezing the market for standard software and large amounts of custom development.
This directly puts software service providers under pressure to transform to AI. Their revenue structures built on licenses, implementation, and custom development could all be affected.
But if the logic that external FDEs can exist long-term holds, then software service providers may also gain new room for survival.
Enterprises can increasingly easily obtain models and code, but they still need someone to understand their evolving business, system architecture, and delivery constraints over the long term. For traditional software service providers, the industry understanding, systems experience, and delivery capabilities accumulated over years at customer sites have exactly the opportunity to be retranslated into agent capabilities.
MingLue Technology is one such example.
Its core business has traditionally focused on marketing intelligence, operational intelligence, data governance, and enterprise software services. Entering the agent era, it began developing agentic services, hoping to further convert its existing software implementation, data governance, and industry service capabilities into custom agent delivery.
MingLue's recent plan to acquire control of Pansoft also continues this line of thinking. Pansoft has long served large group enterprises in petroleum and petrochemicals, coal and power, with a strong focus on custom development and implementation.
Wu Minghui said MingLue plans to first bring FDEs, agents, and related infrastructure into Pansoft, and then have Pansoft extend this capability to its existing customers.
But the key that truly determines whether this logic holds is the leverage ratio of FDEs.
Traditional custom software has always been a labor-intensive business. A project often requires large numbers of engineers on site; the more complex the requirements, the more people are needed, and revenue and headcount have long maintained a strong linear relationship, with limited economies of scale.
For example, Pansoft's revenue grew from 582 million yuan in 2021 to 825 million yuan in 2025, a growth rate of over 40% during the period, but the company has been "working harder and earning less"—its 2025 net profit margin was only 8.09%, down more than 15 percentage points from 2021.
What the agent era seeks to change is delivery efficiency.
Wu Minghui's judgment is that for standard software with clearly defined features, the code itself will become increasingly easy for AI to replicate. Business scenarios where customers haven't yet figured out what to do may instead become the more valuable part of enterprise services.
"If it were just sitting at a computer writing code, there'd be no need to go on site," he said. When engineers come to the customer, the more important work is constantly communicating with the customer to figure out "how the business should actually be run."
According to his description, in the past an implementation person might spend two-thirds of their time communicating with customers and the rest writing code; now, with much of the coding work delegated to AI, FDEs can spend more time on requirements judgment, process decomposition, and result verification.
If the same person can deliver more custom agents and support more customers, the project-based business that once relied on stacking headcount has an opportunity to improve its leverage ratio.
Three
The Payers "Voluntarily Enter the Game"
What AI lowers is not just the development costs of external service providers, but also the barrier for enterprises themselves to enter agent production.
On August 11, All-Weather Technology observed 45 employees at Mengniu's headquarters taking part in an AI application roadshow. They came mainly from business lines such as sales, R&D, supply chain, milk sourcing, and marketing. Over the past three months, Mengniu selected about 200 "AI pioneers" from its 28 first-level business divisions, letting the people most familiar with the business learn to use AI agents first, then return to their own work to find parts that could be transformed by AI.
This arrangement has further evolved into an internal FDE model.
Mengniu plans to establish L1 to L3 AI pioneer certifications. At L3, employees must not only be able to build AI agents, but also understand business needs, develop solutions, and continuously track results. Mengniu's digital technology team has also set up a dedicated AI FDE team internally.
Small and medium-sized enterprises may take an even more direct path. In Yanchi County, Ningxia, Zhang Lifei, general manager of Xixianji, used Alibaba's AI coding tool Qoder to build a livestock farming management system in just over three months. National standards, academic papers, feed formulas, pen-entry records, and frontline farming experience were gradually written into the software, and the system continued to extend along feeding, inventory, slaughtering, processing, and quoting.
The cumulative cost of using Qoder for this system was in the tens of thousands of yuan, whereas the external custom solutions Zhang Lifei had previously been quoted could reach the million-yuan level.
In the past, small and medium-sized enterprises were often stuck between two extremes: off-the-shelf software rarely fit their business perfectly, while custom development was too expensive.
AI coding is starting to push down the development costs in between, bringing some highly customized software into a range that small and medium-sized enterprises can afford.
But it is undeniable that Zhang Lifei's advantages are also quite special.
He has done R&D work and has long been involved in operations, so he knows both how code is implemented and what a sheep eats each day, how feed costs are allocated, and the lowest price a product can be quoted at. Requirements, testing, and usage can all be completed within the company, eliminating the back-and-forth of communicating requirements in traditional custom development.
But this model has its limits too. As the system expands to more departments, testing, permissions, data security, and long-term maintenance become difficult to keep concentrated in one person. Even after AI drives down the cost of writing code, companies still need someone to continuously understand the business, maintain the system, and bear ultimate decision-making responsibility.
Wu Minghui also acknowledged that large enterprises will gradually internalize some FDE capabilities over time. However, as companies move from individual AI use to organization-level AI, they will still encounter issues such as role adjustments, incentive mechanisms, system infrastructure, and cost management, so external forces may still play the initial guiding role.
From this perspective, it will be hard for any of the parties to claim victory in the short term.
Model and cloud providers have underlying capabilities and hope to use FDEs to bring models and AI agents into production; enterprise service providers like MingLue have industry and delivery experience and hope to turn their past know-how into AI agents; large enterprises have begun cultivating internal FDEs, while small and medium-sized enterprises may directly use AI coding to build their own systems.
FDE is therefore more like a boundary that is still moving, and the role itself is far from settled.
Once companies have dozens, hundreds, or even more AI agents, the questions that truly matter are: who will produce these AI agents, who is responsible for maintaining them, and who can turn one business transformation into the next reusable capability.
Around these questions, the enterprise AI market in the age of AI agents may only just be beginning to redivide its roles.
This article comes from the WeChat public account "All-Weather Technology" (ID: iawtmt), author: Zheng Minfang, published by 36Kr with authorization.