ai-product-strategy

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

AI Product Strategy

AI产品策略

Scope

适用范围

Covers

Defining an executable product strategy for an AI/LLM/agent product or AI feature portfolio
Translating AI uncertainty (non-determinism, emergent risks) into an empirical plan with evals + instrumentation
Choosing product form factor (assistant vs copilot vs agent), autonomy boundaries, and a safety/security posture
Producing a strategy pack leaders and teams can use to align and execute

When to use

“Define our AI product strategy / LLM strategy / agent strategy.”
“Prioritize AI use cases and turn them into an AI roadmap.”
“We’re adding AI to an existing product—what should we build and how do we measure it?”
“We want to ship an agent; define autonomy, security, and rollout.”

When NOT to use

You need a long-term product/company vision (use
```
defining-product-vision
```
first).
You need deep competitor research, battlecards, or win/loss (use
```
competitive-analysis
```
).
You need a feature-level PRD/spec/design doc (use
```
writing-prds
```
/
```
writing-specs-designs
```
after strategy).
You’re doing model architecture research, training, or infra-level technical design (delegate to ML/eng).
You don’t yet have a clear problem/ICP hypothesis (use
```
problem-definition
```
/
```
conducting-user-interviews
```
).

涵盖内容

为AI/LLM/Agent产品或AI功能组合定义可落地的产品策略
将AI的不确定性（非确定性、突发风险）转化为包含评估与监控措施的实证规划
选择产品形态（助手、Copilot、Agent）、自主权限边界，以及安全/合规立场
产出可供管理者与团队对齐方向并执行的策略包

适用场景

“制定我们的AI产品策略/LLM策略/Agent策略。”
“对AI用例进行优先级排序并转化为AI路线图。”
“我们要在现有产品中加入AI功能——应该构建什么，如何衡量效果？”
“我们要推出Agent——定义自主权限、安全机制与上线方案。”

不适用场景

你需要制定长期产品/公司愿景（请先使用
```
defining-product-vision
```
）。
你需要深度竞品调研、作战手册或胜负分析（请使用
```
competitive-analysis
```
）。
你需要功能级别的PRD/规格/设计文档（请在完成策略后使用
```
writing-prds
```
/
```
writing-specs-designs
```
）。
你正在进行模型架构研究、训练或基础设施层面的技术设计（请交由ML/工程团队负责）。
你尚未明确问题/目标用户假设（请先使用
```
problem-definition
```
/
```
conducting-user-interviews
```
）。

Inputs

输入要求

Minimum required

Product context (what exists today) + target customer/user + their job/pain
Strategy horizon (default: 3–12 months) + constraints (budget, latency, policy/legal, data access, platform)
Intended AI surface and scope: assistant / copilot / agent; where it lives in the workflow
Success metrics (1–3) and guardrails (2–5), including safety/trust, cost, and latency

Missing-info strategy

Ask up to 5 questions from references/INTAKE.md (3–5 at a time).
If details remain missing, proceed with clearly labeled assumptions and provide 2–3 options (use-case focus, autonomy level, build/buy).

最低必填项

产品背景（现有产品状态）+ 目标客户/用户 + 他们的工作需求/痛点
策略时间范围（默认：3–12个月）+ 约束条件（预算、延迟、政策/合规、数据访问权限、平台限制）
预期AI形态与范围：助手 / Copilot / Agent；在工作流中的嵌入位置
成功指标（1–3个）与防护规则（2–5个），包括安全/可信度、成本、延迟

缺失信息处理策略

从[references/INTAKE.md]中提出最多5个问题（每次3–5个）。
如果仍有信息缺失，基于明确标注的假设继续推进，并提供2–3种可选方案（如用例聚焦方向、自主权限级别、自研/外购）。

Outputs (deliverables)

输出物（交付成果）

Produce an AI Product Strategy Pack in Markdown (in-chat; or as files if requested), in this order:

Context snapshot (decision, users, constraints, why now)
Strategy thesis (value prop, why-now, differentiation, non-goals)
Use-case portfolio (prioritized opportunities with feasibility + risk)
Autonomy policy (assistant→copilot→agent boundaries + human control points)
System plan (build/buy, data plan, eval plan, cost/latency budgets)
Empirical learning plan (experiments, instrumentation, iteration cadence)
Roadmap (phases, milestones, exit criteria, owners)
Risks / Open questions / Next steps (always included)

Templates: references/TEMPLATES.md

以Markdown格式生成AI产品策略包（可在对话中直接输出；若有需求也可生成文件），顺序如下：

背景快照（决策内容、用户群体、约束条件、为何选择现在推进）
策略核心论点（价值主张、推进时机、差异化优势、非目标）
用例组合（按优先级排序的机会点，包含可行性与风险评估）
自主权限策略（从助手→Copilot→Agent的权限边界 + 人工控制点）
系统规划（自研/外购、数据计划、评估计划、成本/延迟预算）
实证学习计划（实验、监控、迭代节奏）
路线图（阶段划分、里程碑、退出标准、负责人）
风险/待解决问题/下一步行动（必须包含）

模板参考：[references/TEMPLATES.md]

Workflow (8 steps)

工作流程（8步）

1) Frame the decision and boundaries

1) 明确决策与范围边界

Inputs: User request + constraints.
Actions: Define the decision to make, strategy horizon, and audience. Decide whether this is for a single feature, a product line, or a platform capability. Write 3–5 explicit non-goals.
Outputs: Draft Context snapshot + scope boundaries.
Checks: You can state “We are deciding X by date Y for audience Z,” and list what’s explicitly out of scope.

输入：用户需求 + 约束条件。
行动步骤：定义要做出的决策、策略时间范围与受众。确定是针对单一功能、产品线还是平台能力。撰写3–5条明确的非目标。
输出物：草稿版背景快照 + 范围边界。
检查要点：需明确表述“我们将在Y日期前为Z受众做出X决策”，并列出明确排除在范围外的内容。

2) Map the user workflow and role shift

2) 梳理用户工作流与角色转变

Inputs: Target user + current workflow.
Actions: Map the workflow steps where AI changes the user’s job. Note “human control points” (where a user must review/approve). Identify failure modes that matter (hallucination, privacy, action mistakes).
Outputs: Workflow notes + role-shift bullets (in thesis or appendix).
Checks: Value is tied to a real workflow step (not generic “AI magic”).

输入：目标用户 + 当前工作流。
行动步骤：梳理AI将改变用户工作方式的工作流步骤。标注“人工控制点”（用户必须审核/批准的环节）。识别关键故障模式（幻觉、隐私泄露、操作错误）。
输出物：工作流笔记 + 角色转变要点（可放在核心论点或附录中）。
检查要点：价值需与真实工作流步骤绑定（而非泛泛的“AI魔力”）。

3) Build a use-case portfolio and prioritize bets

3) 构建用例组合并优先级排序

Inputs: Workflow map + constraints + risk appetite.
Actions: List 6–12 candidate use cases. Score value vs feasibility vs risk. Select the top 1–3 bets and 1 “explore later” bet.
Outputs: Use-case portfolio table + recommendation.
Checks: Each selected bet has a clear user, measurable outcome, and known “must-not-do” constraints.

输入：工作流图谱 + 约束条件 + 风险承受能力。
行动步骤：列出6–12个候选用例。按价值、可行性、风险进行评分。选择前1–3个重点推进的用例，以及1个“后续探索”用例。
输出物：用例组合表格 + 推荐方案。
检查要点：每个选中的用例都有明确的用户群体、可衡量的成果，以及明确的“禁止行为”约束。

4) Define differentiation + “why us / why now”

4) 定义差异化优势 + “为何是我们/为何现在推进”

Inputs: Top bets + assets + market context.
Actions: Draft the strategy thesis: value prop, why-now, and defensible differentiation (data, distribution, workflow integration, UX, trust). Write key assumptions and how you’ll test them.
Outputs: Strategy thesis (copy/paste from template).
Checks: Differentiation is not “we use AI”; it names compounding advantages or unique assets.

输入：重点用例 + 现有资产 + 市场背景。
行动步骤：起草策略核心论点：价值主张、推进时机，以及可防御的差异化优势（数据、分发渠道、工作流集成、用户体验、可信度）。列出关键假设及验证方式。
输出物：策略核心论点（可直接复制模板内容）。
检查要点：差异化优势不能仅停留在“我们使用AI”，需明确指出复合优势或独特资产。

5) Choose form factor and autonomy policy (assistant → copilot → agent)

5) 选择产品形态与自主权限策略（助手→Copilot→Agent）

Inputs: Bets + constraints + safety requirements.
Actions: Decide the minimal autonomy needed for utility. Specify what the system can do, what it can suggest, and what it must never do. Define permission prompts, approvals, logging, and rollback for any action-taking behavior.
Outputs: Autonomy policy table.
Checks: Every action capability has explicit permissions + auditability + rollback.

输入：重点用例 + 约束条件 + 安全要求。
行动步骤：确定实现价值所需的最小自主权限。明确系统可执行的操作、可建议的内容，以及绝对禁止的行为。针对任何可执行操作的行为，定义权限提示、审批流程、日志记录与回滚机制。
输出物：自主权限策略表格。
检查要点：每一项操作能力都有明确的权限要求 + 可审计性 + 回滚机制。

6) Draft the system plan (build/buy, data, evals, budgets)

6) 起草系统规划（自研/外购、数据、评估、预算）

Inputs: Autonomy policy + constraints + data access.
Actions: Choose a strategy-level technical approach (e.g., RAG, tool use, fine-tuning) and a data plan. Define eval strategy (offline + online), quality targets, and cost/latency budgets.
Outputs: System plan.
Checks: There’s a plausible path to meet quality + safety + cost + latency with measurable evals.

输入：自主权限策略 + 约束条件 + 数据访问权限。
行动步骤：选择策略层面的技术方案（如RAG、工具调用、微调）与数据计划。定义评估策略（离线+在线）、质量目标，以及成本/延迟预算。
输出物：系统规划。
检查要点：需有合理路径通过可衡量的评估来满足质量、安全、成本与延迟要求。

7) Make it empirical (experiments + instrumentation + iteration)

7) 制定实证方案（实验 + 监控 + 迭代）

Inputs: Thesis + system plan + assumptions.
Actions: Design experiments/prototypes and a “watch/listen” plan post-launch. Define instrumentation (events/logs), review cadence, and an iteration loop for both utility and risk.
Outputs: Empirical learning plan.
Checks: Every major assumption has a test + metric + owner + timebox.

输入：核心论点 + 系统规划 + 假设。
行动步骤：设计实验/原型，以及上线后的“监控/反馈”计划。定义监控指标（事件/日志）、评审节奏，以及针对效用与风险的迭代闭环。
输出物：实证学习计划。
检查要点：每个重要假设都有对应的测试方法、衡量指标、负责人与时间限制。

8) Roadmap + quality gate + finalize

8) 路线图 + 质量校验 + 最终定稿

Inputs: Full draft pack.
Actions: Create a phased roadmap with milestones, exit criteria, and owners. Run references/CHECKLISTS.md and score with references/RUBRIC.md. Always add Risks / Open questions / Next steps.
Outputs: Final AI Product Strategy Pack.
Checks: A stakeholder can act on the pack without a meeting; trade-offs and unknowns are explicit.

输入：完整的策略包草稿。
行动步骤：创建分阶段路线图，包含里程碑、退出标准与负责人。使用[references/CHECKLISTS.md]进行检查，并通过[references/RUBRIC.md]进行评分。必须添加风险/待解决问题/下一步行动部分。
输出物：最终版AI产品策略包。
检查要点：利益相关者无需额外会议即可基于该策略包采取行动；需明确列出权衡取舍与未知事项。

Quality gate (required)

质量校验（必填）

Use references/CHECKLISTS.md and references/RUBRIC.md.
Always include: Risks, Open questions, Next steps.

使用[references/CHECKLISTS.md]与[references/RUBRIC.md]进行校验。
必须包含：风险、待解决问题、下一步行动。

Examples

示例

Example 1 (AI-first product): “Use

ai-product-strategy

to define strategy for an AI coding assistant for mid-market engineering teams. Constraints: ship a beta in 8 weeks; must not leak proprietary code; budget capped at $X/month.”
Expected: strategy thesis + prioritized use cases + autonomy policy + system/eval plan + roadmap.

Example 2 (AI feature portfolio): “Use

ai-product-strategy

to prioritize AI opportunities for a customer support platform. Decide copilot vs agent, include safety posture, and propose a 2-quarter roadmap.”
Expected: use-case portfolio with 1–3 bets, a clear agency-control policy, empirical plan, and phased roadmap with exit criteria.

Boundary example: “Pick the best LLM provider.”
Response: treat “provider choice” as an input to the system plan; ask for constraints (data, cost, latency, privacy, regions). If the broader product decision is unclear, run this full strategy workflow first.

示例1（AI原生产品）：“使用

ai-product-strategy

为中端市场工程团队的AI代码助手制定策略。约束条件：8周内推出Beta版本；不得泄露专有代码；预算上限为每月$X。” 预期输出：策略核心论点 + 优先级用例 + 自主权限策略 + 系统/评估计划 + 路线图。

示例2（AI功能组合）：“使用

ai-product-strategy

为客户支持平台的AI机会进行优先级排序。确定Copilot或Agent形态，包含安全立场，并提出2个季度的路线图。” 预期输出：包含1–3个重点用例的用例组合、明确的Agent权限控制策略、实证方案，以及带有退出标准的分阶段路线图。

边界示例：“选择最佳LLM供应商。” 回应：将“供应商选择”视为系统规划的输入；询问约束条件（数据、成本、延迟、隐私、区域）。若更宏观的产品决策不明确，请先执行完整的策略工作流。