执行与技术型 PM
与工程和设计协作,可靠地交付 —— PRD、技术素养、敏捷节奏与不会让人意外的发布。
Explainer
执行是产品判断与运营纪律相遇的地方。技术型 PM 不必是房间里最优秀的工程师,但需要足够的系统素养来问出更好的问题、降低交付风险、聪明地权衡范围。执行就是把战略变成一连串可观察的小交付 —— 每一个都是「能让团队学到一件真实的事情」的最小单元。
spec 究竟是什么
spec 不是让工程默默实现的合同。它是一份结构化的论证,记录用户问题、期望结果、被界定的范围、明确的非目标、关键决策、未决风险与上线度量。这份「论证」本身就是交付物 —— 工程读完后应能就权衡进行有理有据的反对。如果 spec 里只有「做什么」,你写的是一张 JIRA 工单。
- 第 1 页结构:问题、用户、当前替代方案、成功度量、范围、非目标。
- 第 2 页结构:关键流程、已作决定、推迟的决定、未决风险。
- 第 3 页结构:上线计划、度量、回滚计划、排序依赖。
- 超过第 3 页的内容属于设计文件或工程 RFC,不属于 PRD。
一页 PRD(精益)
一页 PRD 强迫你写得清楚。任何总工作量在 4 周以内的事项都应使用一页 PRD。只有跨团队倡议、合规相关或「重大押注」项目才用更长的文档。多数团队过度生产文档;字数与执行质量几乎无关。
- 问题:谁在受苦,我们怎么知道,多频繁。
- 结果:我们要换取的指标变化。
- 约束:预算、截止日期、依赖。
- 未决问题:点出来,这样它会被处理而不是被回避。
- 签字:谁说了 yes(DACI 的 Approver)、何时、对哪一版。
路线图格式与各自的取舍
没有「中立」的路线图格式。每种格式都把团队推向不同的对话。日期型 Gantt 偏向承诺;Now/Next/Later 偏向排序;主题型偏向战略对齐;结果型偏向指标变化。选能创造你最需要的对话的格式,并每年重审。
- Gantt:最适合截止日期固定的合规项目;最不适合以发现为导向的产品。
- Now / Next / Later:最适合产品驱动增长;对财务和销售来说信号弱。
- 主题型:最适合做组合式思考;有失去战术细节的风险。
- 结果型:在领导层信任团队选路径时最佳;需要有力的 NSM。
- 混合:许多团队对外用主题—结果,对内用 Now/Next/Later,只有硬截止日期才用日期。
Agile、Scrum、Kanban、Shape Up —— 带着意图选择
方法选择被低估理论化了。Scrum 适用于成熟系统中可预测的功能工作。Kanban 适用于支持流式的工作流。Shape Up 适用于互信稳定团队对赌注的塑形。SAFe 适用于需要跨团队协调的大型受监管组织。没有一种自动正确;每种方法的病灶都被详尽记录。
- Scrum 风险:估算戏剧、仪式疲劳、sprint 目标沦为墙纸。
- Kanban 风险:WIP 爆炸、缺乏明确的优先级地平线。
- Shape Up 风险:对信任的依赖、周期末债务、周期内进展可见度低。
- SAFe 风险:会议负担压垮速度;以仪式替代产品判断。
- 用你团队真实的病灶检验方法,而不是博文热情。
API 以及为何 PM 需要心智模型
API(Application Programming Interfaces)定义系统如何向其他系统请求数据或动作。哪怕非技术 PM 也需要工作模型:REST、GraphQL、RPC、GraphQL 订阅、webhooks。词汇上的差异决定了哪些是容易、哪些是困难。理解 API 表面的 PM 在「暴露什么」「何时引入版本管理」「如何规划合作伙伴集成」上做更好的取舍。
- REST:面向资源(GET /users/123)。多数公开 API。
- GraphQL:由客户端定义查询;灵活,缓存难,后端复杂。
- RPC / gRPC:面向动作(如 CreateInvoice);强类型;通常内部使用。
- Webhooks:服务器对服务器的回调;「在某事发生时通知我」的合适工具。
- 幂等性:可以安全重试且无副作用 —— 计费与集成上不可让步。
数据库、schema,以及它们如何影响路线图
数据库选择长期决定什么便宜、什么昂贵。关系型数据库(Postgres、MySQL)对 join 便宜,对无界规模昂贵。文档库(Mongo、Firestore)对灵活形状便宜,对关系型查询昂贵。键值库(Redis)对缓存与会话便宜,对分析昂贵。能读懂基础 ER 图、并问出「这个查询是有界的吗?」的 PM,会在两年里替团队省下六个月。
- schema 迁移是永久的 —— 每次迁移都是小型 Type-1 决策。
- 读 vs 写:分开扩缩。
- 索引:「突然」性能问题最常见的根因。
- 软删除 vs 硬删除:涉及合规与分析的影响。
- 拿不准时:问「数据放大 100 倍后,这个查询如何扩展?」
队列、重试与幂等性
当系统与系统对话时,网络是不可靠的。队列(RabbitMQ、SQS、Kafka)吸收负载尖峰;重试处理瞬时故障;幂等键防止重复扣款与重复发送。承诺「即时」「实时」用户体验却不理解队列延迟的 PM,会让团队走向无法交付的架构。永远问:这是同步还是入队?
- 同步:用户等待,系统在 <1s 内响应;只适合短小确定性操作。
- 异步:用户得到确认,工作在后台进行;触及外部系统的事情都需要这个。
- 至少一次 vs 仅一次交付:决定幂等性需求。
- 死信队列:消息去那里死掉;应被监控,而不是被忽视。
性能、延迟与尾部行为
中位数延迟会撒谎。用户体验由尾部主导 —— p95、p99、p99.9。一个中位 200ms 但 p99 为 4s 的站点,对 1% 的用户而言是坏的 —— 而那 1% 会是最响亮的。在意 UX 的 PM,在意尾部。
- 中位(p50):一半用户更短;几乎告诉不了你 UX 的事。
- p95:5% 请求更慢;「可接受」的边界。
- p99:1%;在 B2B 中,每个账户每天都会被打到。
- p99.9:千分之一;对高流量产品和 API 至关重要。
Feature flags 与渐进交付
feature flag 是一个运行时开关,能让你把代码上到生产但不暴露给用户。结合按 cohort 的目标定向,它能实现渐进交付:1%、5%、25%、100%,在每一阶段观察指标。把每个有意义的变更视作 flag 化的滚动发布。代价很小;选择权很大。
- 用户可见的任何变更,都先放在 flag 后。
- 始终保留 kill switch —— 不重新部署也能回滚的能力。
- 始终设定 flag 清理时间表;旧 flag 会变成隐形的技术债。
- 把 flag 与指标 watch 配对 —— 退化时自动回滚。
上线检查清单
上线不是「工程合并」。它包含沟通、销售赋能、支持就绪、遥测、成功标准与回滚。多数上线事故并非工程失败,而是清单失败。上线清单是 PM 拥有的、最有杠杆的工件之一。
- 上线前:遥测已验证、成功标准已记录、回滚计划已演练、沟通已起草、支持已培训、销售赋能就绪、beta cohort 反馈已纳入。
- 上线:按指标门控分阶段滚动、on-call 轮班知情、状态页就绪。
- 上线后(T+24h、T+7d、T+30d):指标复盘、定性反馈综合、上线流程回顾。
- 不要在周五或假日前上线 —— 除非你想要一个混乱的周末。
干系人沟通:每周一页
用一页周报替代状态会议。它把信号压缩成 90 秒可读的格式,让风险尽早暴露,并形成「承诺与交付」的公开记录。做得好,可消除 80% 的同步状态对话。
- 顶部:本周 NSM 与头条指标。
- 中部:在做的赌注、状态、信心、阻塞。
- 底部:本周需要的决策(指名 approver、截止日期)。
- 语调:具体而克制。不要营销语言。不要表情包风暴。
- 固定节奏分发(周五 16:00)。可预测性建立信任。
冲突、升级与分歧
PM 在分歧上花的时间多过在共识上的时间。技能不是回避冲突 —— 而是早把冲突摊出来、让它结构化、并由指名决策者解决。多数产品「文化问题」其实是穿了戏服的「决策权问题」。
- 在会议前以书面形式摊出分歧;减少情绪升级。
- 用 steel-manning:在反对前,把对立观点说得比对方更强。
- 用书面形式升级,不要在会议里突袭;指名 approver 决定。
- 决定后 disagree-and-commit;反复翻案具有腐蚀性。
工程信任作为复利资产
PM 与工程之间的信任,是产品中最被低估的资产。被信任的 PM 获得范围弹性、技术支持与更快的交付。不被信任的 PM 得到防御性估时、范围僵硬与神秘的延期。信任建立在无聊的习惯上:偶尔出现在站会、把代码读到能看 PR、为技术债在干系人面前发声、并且永远、永远不在公开场合责怪工程。
- 每季度花半天影子参与一次 on-call 轮班。
- 阅读你没参与过的事故的 post-mortem。
- 公开为技术债与可靠性投入发声 —— 不要让工程独自承担。
- 出现滑坡时,在外部承担责任。工程会在内部回报你。
Framework atlas
Reference cards for each method in this mission
Expand a card for when to deploy it, misuse patterns, sequencing guidance, and (where relevant) shorthand formulas.
Spec format · Amazon (Working Backwards)Amazon PR/FAQ(PR/FAQ)
Write the launch press release and the customer FAQ before building. Forces clarity on the user benefit, not the implementation.
When to use
- New product launches.
- New customer-facing features that need clear positioning.
When not to
- Internal tools, technical foundations.
- Changes too small to merit external messaging.
How to apply
- Write the PR: headline, sub-headline, customer quote, benefit statement, call to action.
- Write the FAQ: 10-15 anticipated customer questions with crisp answers.
- Add the internal FAQ: 5-10 questions about tradeoffs and risks.
- Iterate the doc with cross-functional review until skeptics nod.
- Writing marketing copy instead of customer reality.
- Skipping the internal FAQ — that's where the hard tradeoffs live.
Delivery method · Ryan Singer / BasecampShape Up
Six-week cycles, two-week cooldowns. Shaped bets with fixed appetite. No estimation, no backlog, no sprints.
When to use
- Stable teams with high mutual trust and senior engineers.
- Products without external date pressure.
When not to
- Junior teams without strong shaping skills.
- Date-driven launches (regulatory, hardware, partner).
How to apply
- Shape: senior team frames the bet — problem, appetite (effort budget), boundaries.
- Bet: leadership picks the cycle's bets in a betting table.
- Build: small autonomous teams build for 6 weeks, with hill charts replacing burndowns.
- Cooldown: 2 weeks of slack work, polish, exploration.
- Adopting Shape Up vocabulary without the trust prerequisite.
- Skipping cooldowns under deadline pressure — defeats the model.
Decision rights in execution · Atlassian / IntuitDACI for Specs
Use DACI inside spec docs. Name Driver, Approver, Contributors, Informed for each open decision. Eliminates the 'who said yes?' confusion at launch.
When to use
- Cross-team specs.
- Specs where multiple senior leaders have skin in the game.
When not to
- Trivial specs inside a single team.
How to apply
- List open decisions in the spec.
- Assign DACI roles for each.
- Capture decisions inline as they're made; don't fork into Slack threads.
Progress visualization · Basecamp / Shape UpHill Charts
Visualize work by where it is on a hill (uphill = figuring out, downhill = executing). Replaces percent-complete with a richer mental model.
When to use
- Teams using Shape Up or any flow-based method.
- When percent-done estimates feel meaningless.
When not to
- Hard-deadline date-driven work where percent-complete is contractually relevant.
How to apply
- Define each scope as a movable dot on the hill.
- Move uphill while you're solving the unknowns.
- Move downhill once execution is known.
- Update weekly; stalled dots are the signal.
Engineering decision doc · IETF / open-source cultureRFC (Request for Comments)
Engineering proposal doc shared for cross-team comment before implementation. The eng counterpart to a spec.
When to use
- Architectural decisions, API designs, schema changes affecting multiple teams.
- Changes that lock in long-term constraints.
When not to
- Trivial implementation choices.
How to apply
- Draft the proposal: context, goals, non-goals, design, alternatives considered, tradeoffs.
- Circulate to relevant engineers, PMs, and ops.
- Collect comments async over 3-5 days.
- Final decision documented inline; archive.
Product Psychology
Cognitive biases that distort product decisions
Optimism Bias in Estimates
Engineers and PMs systematically underestimate effort, even when their previous estimates have repeatedly underrun.
Product Risk
Sprint commits slip; quarterly OKRs become aspirational; trust in the team's word erodes over time.
Research Countermove
Track estimate-vs-actual ratios per scorer over time; calibrate. Use reference-class forecasting from past similar work.
Spec Confidence Halo
Once a spec is written and reviewed, its conclusions are treated as more certain than the underlying evidence supports.
Product Risk
Teams continue executing even after countervailing evidence appears, because 'the spec said'.
Research Countermove
Treat the spec as a hypothesis; build mid-implementation review checkpoints; allow scope changes when evidence flips.
Method Worship
Treating Scrum, Kanban, Shape Up, or any other method as inherently virtuous, regardless of whether it solves the team's actual problem.
Product Risk
Process is preserved; the underlying pathology (e.g., unclear priorities) remains unsolved.
Research Countermove
Diagnose the problem first, then pick the method. Re-evaluate annually whether the method still fits.
Status Theater
Substituting visible activity (status meetings, dashboards, ceremonies) for actual progress.
Product Risk
Stakeholders think the team is on track because the meeting cadence is healthy; reality diverges quietly.
Research Countermove
Replace status meetings with weekly written notes. Insist on outcome metrics in status, not activity.
Organizational anti-patterns
When ceremonies look like rigor but aren't
Spec as Implementation Manual
PRD reads like pseudocode; engineering has no room to choose the implementation; design has no room to iterate.
PM is over-controlling, often compensating for low engineering trust.
Write the *what* and the *why*, not the *how*. Save the *how* for engineering RFCs.
The Surprise Launch
Sales finds out about the new feature from the marketing email; support gets bombarded with questions they can't answer.
Launch coordination was treated as an afterthought.
Default launch checklist with sales enablement, support training, and comms drafts as launch-blocking items.
Date-Driven Quality Erosion
Team commits to a date; date approaches; quality bars quietly drop; launch is a buggy mess; team retros, then repeats next quarter.
Date is treated as fixed; scope is treated as fixed; quality is the silent variable.
Make the trade-off explicit: name what would be cut to protect quality; make leadership choose. Don't let 'we'll just push through' be the answer.
Sprint Commitment Theater
Team commits to a sprint; misses by 30%; commits the same way next sprint; nothing changes.
Estimates are political, not actuarial; commits are made to look good in standup.
Track actual vs estimate over time per type of work; use the historical slip factor to right-size commits; make missed commits a planning input, not a performance review.
PM-as-Project-Manager
PM spends 80% of week chasing JIRA tickets, status updates, and sync notes; almost no time in discovery, strategy, or stakeholder alignment.
PM has filled the project-management gap rather than fixing it.
Insist on a project-coordination role or strong eng-lead ownership of execution mechanics. Reclaim PM time for decisions.
Worked examples
Walkthroughs translated from real trade-off rooms
Cutting a 6-page PRD down to 1 page
A first-time PM writes a 6-page PRD for a feature that will take 3 weeks. Engineering review stalls; design feels boxed in.
- Identify the 7 things that *must* be on page 1: problem, user, current alternative, success metric, scope, non-goals, timeline.
- Move flow details into the design file.
- Move technical constraints into an engineering RFC linked from the PRD.
- Move marketing copy into a separate launch doc.
- Result: 1-page PRD that everyone reads in 5 minutes; design has flow ownership; eng has implementation ownership.
TakeawayPRD volume correlates negatively with execution speed. Narrow the doc to argument-shaping content.
Asking the right technical question to save 4 weeks
Spec promises real-time collaborative editing in a doc app.
- PM asks engineer: 'is this synchronous or queued?'. Answer: needs WebSocket and CRDTs (operational transformation), neither of which we have.
- PM asks: 'what's the cost-of-delay if we do polling at 5s intervals first?'. Answer: 95% of the value, 1/4 of the build, 1 week of work.
- Spec is descoped to 'live presence + 5-second sync' for v1, with a clear path to true real-time in v2.
TakeawayA single technical question — synchronous vs queued — saved a month of work and de-risked the launch.
Recovering from a slipped commit with honesty
Team commits to launch by end-of-Q3; mid-quarter, a dependency slips; launch is going to slip 3 weeks.
- PM writes a one-page weekly note disclosing the slip with the new ETA, the cause, and the trade-off being made (quality protected, scope cut, date moved).
- Sets up a 30-minute exec sync with Driver, Approver, key Contributors.
- Presents three options: (a) ship on time with descoped scope, (b) ship 3 weeks late with full scope, (c) ship phased — minimal version on time, full version 3 weeks later.
- Approver picks (c); team and stakeholders align around the new sequence.
TakeawayDisclosing slippage early with options preserves trust; disclosing late with reasons destroys it.
Resources / Case Studies
Curated reading for this mission
Ryan Singer / Basecamp
Basecamp's complete operating model for product delivery: shaping, betting, building, and cooldowns.
The most opinionated alternative to backlog-driven delivery; valuable even if you don't adopt it whole.
Colin Bryar & Bill Carr
Amazon's mechanisms for product development: PR/FAQ, the bar raiser, S-team goals, and operational reviews.
Production-grade operating mechanisms with templates PMs can adapt directly.
Kevin Yien (via Lenny Rachitsky)
Compact PRD template widely adopted by product teams.
Battle-tested format that works for sub-month features; saves PM teams the bike-shed of inventing their own.
Camille Fournier
Engineering management from tech lead to CTO; PMs gain a precise model for what engineering leaders worry about and how they think.
Engineering trust starts with understanding engineering's career, incentives, and pressures.
Will Larson
Engineering management at scale: org design, project planning, and the systemic side of execution.
Helps PMs think about cross-team execution and organizational constraints, not just single-team flow.
Martin Kleppmann
The reference text for distributed systems concepts: consistency, replication, queues, idempotency, scalability.
Skim every chapter. Read the chapters relevant to your product. Single best technical-literacy investment for PMs.
Stripe
Stripe's public API documentation, treated by many as the canonical example of API design quality.
Walk it as a user; you will internalize what excellent API ergonomics feel like.
Google SRE
Google's blameless postmortem template and culture; the gold-standard reference for incident learning.
PMs who run good postmortems compound team learning; bad postmortems compound team fear.
Janna Bastow / ProdPad
The case for Now / Next / Later format and the harm of date-anchored Gantt-style roadmaps.
Foundational reading for moving teams off date-driven roadmap pathology.
Patrick Lencioni
Team-dynamics model: trust, conflict, commitment, accountability, results — and what breaks each.
PM execution problems are often team-dynamics problems; this book gives vocabulary to the patterns.
Patterson, Grenny, McMillan, Switzler
Tactical guide to high-stakes conversations: psychological safety, surfacing dissent, mutual purpose.
PM job is mostly conversation; a 5% improvement in conversation quality compounds across the year.
Lenny Rachitsky
Practitioner essays on team operating systems, launches, and execution rhythms.
Current real-world examples to balance the canonical books.