FDE (Forward Deployed Engineer) has become the "new darling" of the AI world. In early October, Anthropic announced it would spend $100 million to launch the "Claude Forward Institute," aiming to train 10,000 FDEs by the end of 2027. Major Chinese companies are likewise pouring big money into talent competition: ByteDance offers Doubao FDEs annual packages of up to 1.05 million yuan, and Moonshot AI's Kimi enterprise FDEs up to 1.09 million yuan...
On August 31, the Ministry of Industry and Information Technology issued the "Notice on Launching a Special Action to Cultivate AI Application Service Providers," proposing to "encourage service providers to build FDE teams and embed themselves at customer sites to ensure AI scenarios land in practice." This is the first time FDE has been written into a national-level policy document. The AI Research Institute of the China Academy of Information and Communications Technology stated, "This notice marks FDE's transition from industry exploration to policy deployment, making it a key lever for enhancing AI application service capabilities."
On this new track, some companies have already achieved real-world results. Lenovo recently published a case in which, through the FDE model, it built a complete AI-native application for a large domestic home appliance manufacturer, delivering two AI agents in three months and reducing finished-goods inventory by 70%. At the same time, Lenovo has established a dedicated FDE organization whose core purpose is to participate early in customers' business planning, driving a shift from "responding to customer needs" to "defining customer value."
Behind the talent war, FDE is being pushed wholesale to the industry front line as the "critical link" in enterprise AI delivery.
FDE fills the gap in AI delivery; three mainstream models each have pros and cons
Industry insiders generally agree that FDE is the key role in bringing AI to thousands of industries. In a research report, China Galaxy Securities described the FDE role more concretely — its core is solving the "last mile" of AI application deployment, essentially a system for the deep delivery of AI applications and AI agents.
In this system, FDEs must "embed" themselves in the customer's business site, untangle complex approval workflows, clean up dirty data scattered everywhere, and fine-tune AI agents in real business systems. What they deliver is no longer a server with software installed, but a "digital employee" that can actually run within the customer's business and do work on people's behalf.
A large number of companies are accelerating their shift toward this model. Gartner analyst Alex Coqueiro predicts that by the end of 2026, more than 85% of technology vendors will make FDE programs their core method for delivering AI.
Currently, the global FDE track is roughly divided into three approaches. Model vendors take the "heavy customization" route: OpenAI, Anthropic, ByteDance and others have chosen to get personally involved, dispatching engineers in organized teams to customer sites, trading high investment for high customer lock-in. For customers, the cost of this model is that the deeper the system is embedded, the higher the switching cost, which may ultimately evolve into long-term dependence on a single vendor.
Consulting firms choose the "replicable universal framework" route: consultancies and IT service providers such as Accenture, Deloitte and IBM incorporate FDE into their existing consulting and implementation business frameworks, attempting to standardize and productize it. But universal frameworks struggle to fit customers' highly non-standard business underpinnings; once beyond standard scenarios, FDE easily degenerates into traditional per-person-day billed consulting.
Cloud vendors choose the "ecosystem certification" route: rather than maintaining massive FDE teams themselves, they release capability certification systems and recruit third-party service providers. For example, at the Apsara Conference, Alibaba Cloud, together with Qianwen Office, launched a public cloud ecosystem partner FDE system, and Tencent Cloud launched an FDE engineer certification. The core of this route is controlling the power to define standards, solving the problem of "how to scale quickly." But its pain point is equally prominent: the customer's actual experience often depends on the specific service provider's capabilities rather than the platform standard.
FDE success is not "project delivery" but "whether assets are accumulated for the customer"
"If FDE success is defined merely as 'project delivery,' then what the customer buys is nothing more than 'professional services.'" Gartner analyst Mukul Saha offered this sharp judgment on the current state of the industry. The three routes mentioned above each have their own business logic, but none answers this question: when the vendor leaves, can the business logic and experience proven on-site become the customer's own assets?
Lenovo's "Tianqiao AI + FDE" dual-wheel drive may offer a different answer. According to reports, on one side, with Tianqiao AI as the foundation, capabilities such as model integration, agent orchestration, skill governance and runtime monitoring are standardized in the Agent Workshop, providing enterprises with a reusable AI production environment; on the other side, through the FDE mechanism, engineers are deployed early to customer sites to work with business teams on scenario exploration, process restructuring and agent fine-tuning, adapting the general platform to the customer's specific business form.
In the case of the large domestic home appliance manufacturer published by Lenovo, the entire delivery process was divided into three steps: the FDE team deployed early to the customer site and worked with the business side to identify two high-value scenarios — product lifecycle and inventory challenges, and inaccurate sales forecasting and demand planning; leveraging the Agent Workshop, POC validation was completed within 3 weeks, locking in viable agent directions; and within three months, two agents for inventory optimization and planning were rapidly delivered.
The business logic proven through this model was accumulated as the customer's own "Agent Workshop," giving the customer the ability to iterate independently going forward. The scenario exploration conclusions drawn on-site become templates for agent orchestration; the results of process restructuring become rules for skill governance. Through continuous operation, in addition to reducing finished-goods inventory by 70%, the enterprise also achieved a 45% reduction in raw material inventory, a 38% reduction in order lead times, and a 50% reduction in stagnant product inventory.
The success of this case validates the value of Lenovo's "Tianqiao AI + FDE" dual-wheel drive. It is understood that Lenovo is further strengthening front-end and back-end coordination for FDE. Its newly established FDE organization, building on the integration of the solutions expert team, has also set up a supporting resource pool and established a committee on the R&D side, focusing on interpreting customers' intelligent transformation strategies, defining value scenarios, designing solutions, and accumulating agent assets, driving the transformation from customer needs to business outcomes.
Industry insiders say that Lenovo's move is, in essence, an exploration of the transformation of delivery models in the AI era — using organizational mechanisms to convert the non-standard experience gained through hard-won front-line work into reusable, standardized assets on the back end, thereby pushing the FDE model deeper into the industrial heartland.