Customer Wants Full Autonomy(客户要完全自动化)
知识库位置:
FDE Knowledge Base/09 Field Guides/06-Customer-Wants-Full-Autonomy.md
Situation(情景)
Section titled “Situation(情景)”客户说:
我们不想要 Copilot(智能副手),我们要全自动 Agent(智能体)。
What It Usually Means(通常意味着什么)
Section titled “What It Usually Means(通常意味着什么)”客户通常追求的是:
- 更高效率
- 更低人力成本
- 更少人工步骤
但:
Full Autonomy(完全自主)是实现方式,不是业务目标。
Questions to Ask(该问什么)
Section titled “Questions to Ask(该问什么)”- 哪些动作需要自动化?
- 每个动作出错的损失是多少?
- 是否可逆?
- 是否涉及金钱、删除、外部发送或审批?
- 当前人工流程错误率多少?
- 谁对最终结果负责?
- 是否需要审计?
Risk Classification(风险分级)
Section titled “Risk Classification(风险分级)”Low Risk(低风险)
Section titled “Low Risk(低风险)”- Search(搜索)
- Summarize(总结)
- Classify(分类)
Medium Risk(中风险)
Section titled “Medium Risk(中风险)”- Create Internal Draft(创建内部草稿)
- Add Note(添加备注)
- Update Non-critical Field(更新非关键字段)
High Risk(高风险)
Section titled “High Risk(高风险)”- Refund(退款)
- Send External Communication(发送外部沟通)
- Delete(删除)
- Payment(支付)
- Account Changes(账户变更)
Recommended Architecture(推荐架构)
Section titled “Recommended Architecture(推荐架构)”不要把 Autonomy(自主等级)做成:
Off / On(开/关)建议建立等级:
Level 0
Section titled “Level 0”AI Recommendation only(仅 AI 建议)。
Level 1
Section titled “Level 1”User-triggered Action(用户触发的动作)。
Level 2
Section titled “Level 2”Auto-execute Low-risk Actions(自动执行低风险动作)。
Level 3
Section titled “Level 3”Conditional Autonomy with Policy(基于策略的条件自主)。
Level 4
Section titled “Level 4”Broad Autonomy under Explicit Boundaries(明确边界内的广泛自主)。
Safer Path(更安全的路径)
Section titled “Safer Path(更安全的路径)”Recommend(建议)↓Human Confirm(人工确认)↓Auto Low-risk(低风险自动化)↓Expand Boundaries(扩大边界)根据生产证据逐步升级。
What NOT to Do(不要做什么)
Section titled “What NOT to Do(不要做什么)”不要为了满足客户一句:
我们要 Agent
就直接给:
- Admin Credentials(管理员凭证)
- Unlimited Tools(无限工具)
- No Approval(无审批)
- Unlimited Steps(无限步数)
Key Principle(关键原则)
Section titled “Key Principle(关键原则)”Autonomy should be earned through evidence.
(自主权应通过证据逐步获得。)