AI's Second Half: Why Delivery Beats Parameters — The FDE Role in National Policy

01. Is It a Technical Problem?

After 12+ years in enterprise IT, the author hears two recurring complaints: "The model I tried performed just okay" and "The demo ran well but failed on real workflows." Many treat this as a technical shortfall — insufficient models, data, or infrastructure. The author argues the real gap is not models (which are abundant) but people who can embed models into business processes and translate them into measurable results on business reports. The industry calls this role the Frontier Deployment Engineer (FDE), a title that in 2026 appeared for the first time in a national-level government document (a "red-header" file).

02. Interpreting Document No. 414

The Ministry of Industry and Information Technology (MIIT) issued the "Notice on Carrying Out the Special Action for Cultivating AI Application Service Providers" — known in the industry as "Document 414." A key sentence reads: "Encourage service providers to build Frontier Deployment Engineer (FDE) teams, root them at user sites, and guarantee scenario implementation." This elevates a role previously seen only in job postings and industry salons to national policy status.

The document sets quantitative targets: by end of 2026, the national service-provider resource pool must contain 2,000 firms; by end of 2027, no fewer than 3,000. It breaks core service-provider work into four areas — consulting & planning, delivery & implementation, operations management, and security governance — only one of which touches "pure algorithms." It also establishes a "service team" mechanism: each team led by one service provider with at least two upstream/downstream partners, with at least 310 teams across China's 31 provincial-level regions. The policy signal is clear: the push is not for "whose model is stronger" but for "who can make models run inside enterprises."

03. The Model-Centric Narrative Must End

For the past two years, AI discourse has been dominated by parameters, leaderboards, and launch events. The louder the launch, the quieter the actual deployment. Models keep getting stronger, yet few survive the transition into workshops, processes, reports, and daily operations. The bottleneck is the delivery phase, long neglected. A model that cannot land is worth zero. This mirrors the ERP era 20 years ago: the hard part was not the software but reshaping business processes, connecting data, and getting people to adopt it. Today's AI challenge is the same, only renamed.

04. Delivery Philosophy: Core Is Value Delivery

Industry insiders describe FDE as a "delivery philosophy." The author agrees but adds that it is not merely a job title or a craft — it is an entire value system for viewing delivery. However, philosophy must be actionable. The core is value delivery : delivery is judged not by how many features are shipped but by how much business metrics improve. Before delivery, set baselines, targets, and define "what done looks like." After delivery, track metrics, trends, and whether the business was genuinely improved. A true FDE team dares to tie its own assessment to the client's business KPIs; otherwise it is just rebranded on-site outsourcing.

OpenAI's FDE team defines the mandate as "leave with product" — when you leave the client site, you leave behind a reusable product, not a slide deck. Palantir states: "FDE exists not for demos but for real business results. Demos are for show; business results are for operating reports. The sole reason for FDE existence is to move the client's key business metrics." This echoes the ERP playbook: set metrics first, verify, then account — now upgraded from "accounting" to "accounting for business metrics."

05. Value Delivery Is the Target; Four-Ring Integration Is the Method

Value delivery requires a concrete method. Officially termed "Frontier Deployment Engineering," it comprises four integrated rings: Business Reshaping, AI Engineering, Evaluation & Verification, Production Operations . In plain language: Business Reshaping means redesigning processes, not just adding AI to old ones; AI Engineering means production-grade deployment, not just calling an API; Evaluation & Verification means data-driven proof, not "feels good"; Production Operations means continuous running, not "launch and done." Two preconditions: multi-party co-creation (service provider and client work together) and agile iteration (small fast steps, not big waterfall). The structure: value delivery is the target; four-ring integration is the method. Miss one ring and you miss the target.

06. Six-Layer Value Metrics to Quantify Delivery

Quantification uses six layers:

Financial Value: cost reduction amount, unit cost reduction rate, payback period.

Operational Value: processing time, throughput, on-time delivery rate, equipment OEE, downtime.

Quality Value: first-pass yield, defect rate, scrap/rework rate.

Growth Value: conversion rate, average order value, incremental revenue.

Adoption Value (AI-specific): weekly active usage rate, automation rate, manual review rate, false positive/negative rate.

Reuse Value: number of reusable assets沉淀, cost reduction when replicating to next scenario, delivery cycle time.

Metrics must be used in three steps: (1) before delivery, establish baselines; (2) during delivery, monitor trend direction; (3) after delivery, retain operations — metrics must keep turning after the team leaves. Quantifying value makes delivery auditable and distinguishes it from "one-off project closure."

07. The Name Determines Survival

Naming FDE is not wordplay. Names direct resource flows: talent, capital, and policy incentives. "On-site engineer" attracts outsourcing mindset — finish a project and zero out. "Frontier Deployment Engineer" attracts people who want to deliver lasting assets. A single word difference separates two tribes and defines the industry watershed.

08. AI's Second Half: Watch Delivery, Not Parameters

Legitimizing FDE is essentially legitimizing AI deployment. The past two years were model-driven narrative; now the narrative must return to value delivery. The national document names not a parameter but a delivery mode. This signals China's AI is moving from capability demonstration to large-scale value delivery. The answer is in the document; the question is who can execute.

Closing Advice for Three Audiences

Practitioners: FDE is the most certain career track for the next few years — close your skill gaps now.

Service Providers: Delivery capability is your moat; build FDE teams to secure a spot in the resource pool.

Clients (Buyers): Judge an FDE team by its stance on business metrics. Does it dare tie its assessment to your KPIs? If yes, it's a real FDE.

Ultimately, FDE is not a role but a delivery philosophy whose core is value delivery: targeting real business scenarios with verifiable business value as the goal. Understanding value delivery means understanding the entire answer to AI deployment.