Guotai Haitong: FDE and Harness create a commercial flywheel to support the commercialization of AI.

Guotai Haitong released a research report stating that FDE is responsible for in-depth business scenarios and consolidating industry experience, while Harness ensures the stable execution and reuse of capabilities of intelligent agents. The collaboration between the two is expected to drive enterprise AI towards more efficient and replicable large-scale deployment. The bank believes that the transition of enterprise AI from Demo to Production will lead to an industrial value shift from simple model capabilities to "business understanding + engineering delivery + Agent Runtime + platform reuse." It is recommended to focus on vendors that possess industry scenario know-how, FDE delivery capabilities, and Agent platform consolidation abilities. Guotai Haitong's main points are as follows: FDE is reorganizing the delivery model of enterprise AI from product to production. FDE (Forward Deployed Engineer) is not a traditional pre-sales or implementation position; rather, it recombines capabilities such as business understanding, scenario recognition, prototype verification, system integration, production deployment, and effect evaluation, linking client business sites with product development. Once AI enters the core processes of enterprises, it requires adaptation to differentiated data, systems, permissions, rules, and organizational processes, leading to a deeper customization of requirements; at the same time, AI Coding and Agent tools significantly reduce the costs of customized development, making the previously economically weaker "heavy delivery" model viable for scaling. The true scalability of FDE does not lie in continuously increasing personnel but in consolidating on-site experience into Skills, connectors, industry templates, and platform capabilities, which continuously decreases the human resource investment per client. Harness has become the essential infrastructure for the evolution of Agent capabilities, shifting model competition from isolated capabilities to system engineering. Essentially, Harness is an Agent Runtime encapsulated outside the foundational model, responsible for context management, memory management, tool invocation, state maintenance, permission control, result verification, exception recovery, and long-term task management, organizing a single model invocation into a continuously running Agent. As Agents progress from single-round Q&A to multi-step, cross-system, and long-duration task execution, the accumulation of errors, state drift, tool misuse, and security risks increases, thereby enhancing the importance of Harness. In the future, the competition for Harness is expected to shift from how many components are integrated to enhancing real task outcomes with reasonable Runtime costs, with Cost per Successful Task likely becoming a core indicator reflecting Agent productivity more closely than single-token pricing. The combination of FDE and Harness forms a commercialization flywheel for AI, moving enterprise AI from human replication to capability replication. FDE addresses business understanding and implementation, while Harness resolves stable execution and capability consolidation. Together, they continuously encode the latent business knowledge, process rules, and engineering experience from enterprise sites into executable capabilities for Agents. FDE enters client sites to acquire high-value business knowledge, which Harness then converts into Skills, Tools, Workflows, Ontologies, and operational rules; the production environment further generates real tasks, failure cases, and user feedback, feeding back into the iteration of FDE and Harness. As common experiences continue to consolidate, the deployment for new clients shifts from zero development to configuration and reuse, ultimately forming a commercialization loop where deployment speed increases, marginal delivery costs decrease, and revenue grows faster than the labor growth in delivery. Risk Warning: The risks of enterprise AI commercialization and scaling deployment progress falling short of expectations; the risks of downstream client AI budgets, scenario expansion, and ROI fulfillment not meeting expectations; and the risks of intensified industry competition and rapid changes in technological pathways.