Summary of Key Points
The current bottleneck in the implementation of AI has shifted from a lack of models to the adaptation of scenarios and the transformation of business processes. Domestic large-scale models, such as Yuezhi Dianmian Kimi K3 and Alibaba Qwen3.8-Max, have parameters exceeding one trillion, but companies face practical issues such as data access rights, process integration, and responsibility allocation. Creating demos is easy, but they often go unused (this is common with monthly or weekly updates of AI solutions). At this point, FDEs (Frontier Deployment Engineers) become a critical role: they bridge the gap between technology and business by validating scenarios in advance and coordinating multiple teams to move AI from pilot projects to large-scale, practical applications. Tech giants, with their comprehensive capabilities, are better positioned to lead the adoption of the FDE model.
New Bottlenecks in AI Implementation
The problem is no longer the scarcity of models; there is an abundance of them (two models with over two trillion parameters have been released within a week). However, new challenges have emerged:
- Where is the data? Companies often have a mix of old and new systems, with data scattered across different departments. Even if the physical infrastructure is unified, issues like responsibility boundaries for accessing data still need to be addressed (for example, who can access financial data?).
- Do we need to change processes? Business personnel have been using traditional methods for years; can AI-generated results be directly integrated into their work? And who is responsible if something goes wrong?
- Demos don’t reflect real-world scenarios: Creating a demo by gathering information and adding a natural language interface may take weeks, but it might not be used in production because the actual business pain points (such as real-time data updates or compliance requirements) are not addressed.
These “details” are the real obstacles to the successful implementation of AI.
What is an FDE? Why Are They in High Demand Now?
FDEs are not a new role (Palantir had them more than a decade ago), but they have gained prominence in the era of large-scale models:
- Role Definition: They act as intermediaries between technology and business, understanding both AI capabilities and the specific needs of businesses.
- Why Are They More Needed in China? The level of digitalization varies significantly among Chinese companies—some are highly digitized, while others need to catch up. Within the same company, new systems coexist with outdated software. AI cannot ignore these historical constraints, and FDEs help bridge these gaps.
- OpenAI’s Commitment: In May 2026, OpenAI established a Deployment Company, making FDE services a formal business line, indicating global recognition of this role.
How Do FDEs Work?
The involvement of FDEs changes traditional project processes, focusing on “early experimentation and accelerated collaboration”:
- Reversed Process: In traditional projects, 20% of the time is spent defining requirements, and 80% on development; with AI, 80% is spent on validation (creating a minimal viable product for business testing), and only 20% on development. For example, CCTV Sports’ World Cup AI assistant was launched within a week for user testing and continuous improvement.
- Collaborative Teams: FDEs bring together business, data, and IT personnel from the client, as well as vendor products, algorithms, and engineers. The Peking University International Hospital project completed seven patient service scenarios in three weeks due to direct communication, reducing information loss.
- Visible Results: The AI application at CNPC Kunlun Digital Intelligence is the first 24/7-running AI within their system, proving that AI can be integrated into core business processes.
How Can the FDE Model Be Sustainable?
For the FDE model to be sustainable, three aspects need to be considered:
1. Client Perspective: The value must be tangible, such as time savings and risk reduction—not just the number of AI projects launched, but whether business outcomes improve (for example, providing 24/7 services for 1.5 million patients at Peking University Hospital).
2. Vendor Perspective: Revenue cannot rely on a one-time project fee; it must come from ongoing services (model usage, computing power, maintenance, etc.). Alibaba Cloud’s FDE service generates revenue through continuous technical support.
3. Product Development: The experience gained by FDEs should be reflected in product improvements. For example, if a client encounters an issue with data access rights, it can be fixed as a standard feature in future products, avoiding repeated problems (this is what Palantir calls “human reverse propagation”).
Payment Models Are Changing
The payment model is shifting from a one-time fee to annual services, and in the future, it may evolve to pay based on outcomes (for example, reducing failure rates).
The Ultimate Purpose of FDEs
FDEs are like the electric revolution of the 19th century: initially, factories simply replaced steam engines with electric motors, but productivity increased only when each machine was equipped with an electric motor and processes were reorganized. Similarly, the true value of AI lies not in the day a model is released, but in the moment it redefines business processes according to logical business logic. FDEs play a key role in this transformation, ensuring that AI becomes an integral part of business operations, rather than a standalone tool.
Conclusion
FDEs are more than just engineers; they are the “bridges” that enable the successful implementation of AI in real-world applications. As tech giants transform FDE practices into standardized products, the goal of using AI to empower various industries will be truly achieved.