这是三层台阶的 B 层 · 核心流程。如帕兰蒂尔一般，由一位 FDE（Forward Deployed Engineer）驻场，深入贵公司最关键、最不容差错的业务流程，定制开发贴合实际的工作流智能体。报价不再采用按周期固定总价，而是「每日入驻基础单价 × 智能体化难度系数」——为“自动化的真实难度”提供清晰定价。This is layer B · core process. As Palantir does, a Forward Deployed Engineer (FDE) works on-site, goes deep into your most critical, error-intolerant processes, and builds workflow agents that fit reality. Pricing is no longer a fixed cycle total — it is daily base rate × agentification difficulty coefficient, giving a transparent price for how difficult the work truly is to automate.
将传统的“按周期固定总价”拆解为两个可解释的部分：一个公开、固定的基础单价，乘以一个有标准、可复核的难度系数。We decompose the traditional fixed-cycle total into two explainable parts: a public, fixed base rate times a standardized, auditable difficulty coefficient.
¥6,000 / 人·天，对应难度系数 1.0 的最简智能体工作。透明、固定，不藏价。¥6,000 / person·day, the simplest agent work at coefficient 1.0. Transparent, fixed, nothing hidden.
由 Opus 4.8 按统一的 ADC 五维标准评估，1.0–3.0，分 L1–L5 级。同一把尺，谁来量都一样。Assessed by Opus 4.8 on a unified 5-dimension ADC standard, 1.0–3.0, tiered L1–L5. One ruler for everyone.
系数与人天均给出五维打分依据；随真实交底材料复核，先小范围验证再行扩展，不做一次性的宏大承诺。Both coefficient and person-days come with a 5-dimension rationale, re-checked against real materials. Validate at small scale first, no all-at-once promises.
一项业务工作"交给智能体"到底有多难？我们把它拆成五个工程维度，每维 1–5 分。这不是拍脑袋——每一分都对应明确的工程判据，让难度可被讨论、可被复核。How hard is it to hand a piece of work to an agent? We break it into five engineering dimensions, each scored 1–5. Not guesswork — every point maps to an explicit engineering criterion, so difficulty can be discussed and audited.
评估方：Opus 4.8。系数为按统一标准给出的工程难度刻度，非对客户行业的价值评判；同一模块在不同企业的数据就绪度（D3）不同，系数会相应微调。Assessed by Opus 4.8. The coefficient is an engineering-difficulty scale on a unified standard, not a value judgment on the industry; the same module's data readiness (D3) varies by company, so the coefficient is fine-tuned accordingly.
下表按统一标准给出 8 个场景的难度系数与实际零售单价（基础单价 ¥6,000/人·天）。每个场景的五维打分与"AI 到底做什么"见表下卡片。Below are 8 scenarios scored on the same standard, with retail day rates (base ¥6,000/person·day). Each scenario's 5-dimension scores and "what the AI actually does" are in the cards underneath.
| 业务场景Scenario | 行业Industry | 难度 ΣΣ | 分级Tier | 系数Coef. | 实际零售单价 / 人·天Retail / person·day |
|---|---|---|---|---|---|
| 美国外观专利附图审查报告US design-patent drawing review | 涉外知识产权Foreign IP | 15 | L3 | 2.0 | ¥12,000 |
| 专利查新检索报告Patent prior-art search report | 知识产权IP | 18 | L4 | 2.3 | ¥13,800 |
| 方案设计文档自动化Solution-design doc automation | 工业装备Industrial equip. | 11 | L2 | 1.6 | ¥9,600 |
| 方案 / 报价文档自动化Proposal / quote doc automation | 工业装备Industrial equip. | 13 | L3 | 1.8 | ¥10,800 |
| 获客与市场情报Lead-gen & market intel | 工业装备Industrial equip. | 16 | L3 | 2.1 | ¥12,600 |
| 销售助理 BotSales-assistant bot | 工业装备Industrial equip. | 14 | L3 | 1.9 | ¥11,400 |
| 知识标准化与销售赋能培训Knowledge standardization & enablement | 工业装备Industrial equip. | 19 | L4 | 2.4 | ¥14,400 |
| 技术研发 + CAD / 3D 生成R&D + CAD / 3D generation | 工业装备Industrial equip. | 23 | L5 | 2.8 | ¥16,800 |
智能体读取外观专利附图（多模态视觉），对照 USPTO 制图规范（37 CFR 1.84 / MPEP 1503）逐项审查——视图齐全性、虚实线、剖面线与阴影、各视图一致性、附图编号——产出"问题清单 + 整改建议"审查报告，供代理师复核。The agent reads the design-patent drawings (multimodal vision) and checks them against USPTO drawing rules (37 CFR 1.84 / MPEP 1503) — view completeness, solid/broken lines, shading, cross-view consistency, figure numbering — producing an "issue list + fixes" review report for the attorney.
输入技术交底书，跨 USPTO / EPO / Google Patents / CNIPA 四通道做 28 轮递进检索，产出检索要素表、技术特征对比矩阵、X-Y-A 文献分类与创造性评述（对标智慧芽 PatSnap）。多源编排叠加新颖性 / 创造性的法律判断，高风险。From a disclosure, run 28 progressive search rounds across USPTO / EPO / Google Patents / CNIPA, producing a feature table, comparison matrix, X-Y-A classification and inventiveness commentary (a PatSnap-class report). Multi-source orchestration plus novelty/inventiveness legal judgment — high stakes.
设备模块化组合、工艺已定型，从配置参数 + 组件功能库自动拼装设计 / 开发 / 操作文档初稿，助理从"照旧版逐字改"变成"审核 AI 初稿"。范本现成、参数化拼装，最容易起步。Equipment is modular and the process is fixed; assemble design / dev / operation doc drafts from config parameters + a component-function library, turning "edit the old version line by line" into "review the AI draft." Templates ready, parametric — the easiest start.
一个谈判周期产生 6+ 版方案，自动同步价格 / 配置 / 文本三处，并做全文一致性校验，消灭"改了以为改了、隐患到合同执行才暴露"。跨文档一致性逻辑 + 报价勾稽。A negotiation cycle spawns 6+ versions; auto-sync price / config / text across all three and run a full-document consistency check, killing the "thought I changed it" risk that surfaces only at contract time. Cross-doc consistency + quote reconciliation.
从已成功案例横向发散找同行（做通一家→系统找到同行业另 9 家）、显性企业背调 + 隐性需求挖掘（从生产环节反推需求）+ 细分市场报告。多源爬虫 / API，叠加隐性推理与销售背书。From a won case, fan out to peers (win one → systematically find 9 more in the industry), explicit company due diligence + implicit-need mining (infer needs from production), plus niche market reports. Multi-source scraping/API, plus implicit reasoning and sales validation.
拜访前情报包、拜访后录音转写挖需求、拜访中实时专业问答、新人陪练。多模态（音频）+ 实时检索；输出为建议性、低风险，所以校验维度（D4）较低。Pre-visit briefs, post-visit audio transcription to mine needs, in-visit real-time expert Q&A, and rookie role-play. Multimodal (audio) + live retrieval; output is advisory and low-stakes, so D4 is low.
把 27 年 / 400+ 工艺的隐性经验，沉淀为按细分行业的"标准产品 + 竞争优势 + 价格策略"知识卡片库 + 培训 Bot。知识高度隐性（D2=5）、需创始人逐项审核在环，ROI 最高、门槛也最高。Distil 27 years / 400+ processes of tacit experience into per-niche "standard product + edge + pricing" knowledge cards + a training bot. Knowledge is highly tacit (D2=5) and needs founder sign-off in the loop — highest ROI, highest bar.
对接 CAD 内核（CadQuery 类）做参数化建模 / 图纸校验，乃至高分子 / 分子设计（AlphaFold 类）的情报跟踪。生成式工程 + 物理正确性难校验（D4=5）+ 深度系统耦合（D5=5），前沿 / 研究级，先做可行性评估不前期承诺交付。Drive a CAD kernel (CadQuery-class) for parametric modeling / drawing checks, up to polymer / molecular-design (AlphaFold-class) intel tracking. Generative engineering + hard-to-verify physical correctness (D4=5) + deep coupling (D5=5) — frontier; we scope feasibility first, no upfront delivery promise.
凡涉及法律判断、新颖性 / 创造性认定、答复定稿、合同与报价定价——AI 只做信息梳理与检查清单，结论由代理师 / 工程师 / 管理者判定。For legal judgment, novelty/inventiveness calls, final responses, contracts and pricing — AI only organizes information and builds checklists; the conclusion is the professional's call.
FDE 在客户本机 / 内网驻场开发，交底书、案件材料、图纸全程私有化处理，符合工业与专业服务的安全底线。The FDE develops on the customer's own machine / intranet; disclosures, case files and drawings are handled privately throughout — meeting industrial and professional-services security bars.
先以一个低系数、高确定性的模块完成验证、建立信任，再逐步推进至高系数模块；系数与人天随真实材料复核，不一次性承诺覆盖全部六个场景。Validate and build trust with a low-coefficient, high-certainty module first, then progress to higher-coefficient ones; coefficient and person-days are re-checked against real materials — no commitment to cover all six scenarios at once.
我们将先按 ADC 标准提供难度系数与人天估算，再确定驻场方案与报价。We'll first provide a difficulty-coefficient and person-day estimate on the ADC standard, then finalize the on-site plan and quotation.
