The Skill Stack: What PMs and CPOs Must Learn Now

The stack moved

For twenty years the PM skill stack was stable: write specs, run rituals, coordinate execution, communicate up. Coordination was the scarce thing, so people who coordinated well got promoted. That stack quietly stopped paying. Agents absorbed the mechanical layer, the translator function died with it, and the value moved in two directions at once: up into judgment, and out into expression.

This chapter is the map I wish I could hand every PM and CPO I have worked with: twelve skills in five layers, each with what it is, why it matters now, a falsifiable self-test, and the fastest acquisition path I know, on this site and off it. It is a long chapter. It is meant to be returned to, not consumed.

The short version

The skill stack has five layers. Judgment: problem selection, decision quality, taste. Expression: storytelling, writing for machines, prototyping. Systems: evals, agent orchestration, signal architecture. Economics: unit economics, pricing. Leadership: coalition building and killing things well. IC PMs live in expression and systems while building judgment; CPOs are paid for judgment, economics, and leadership. The two highest-ROI skills to start with are evals (rarest, most employable) and storytelling (multiplies everything else). Every skill below has a self-test; run them all, and spend the next two quarters on your two worst scores, not your two favorites.

Layer 1: Judgment

The layer agents cannot reach, and the one everything else exists to serve.

1. Problem selection

Choosing what deserves to be built at all, when building is no longer the expensive part. The cost of being wrong did not fall with the cost of execution; it rose, because you can now be wrong faster and at greater scale. I wrote about this in the cost of being wrong(coming Jun 18).

Self-test: list the last five things your team shipped. For each, can you state the evidence that made it the best available use of the team at the time? Not the justification, the evidence.

Acquire it: run a real opportunity solution tree (Your First OST), then read Teresa Torres, Continuous Discovery Habits, and practice the weekly cadence in the weekly discovery playbook.

2. Decision quality

Treating decisions as bets with odds, keeping score, and separating decision quality from outcome quality. Most product orgs have zero memory of their own decisions, which means zero learning loop. The deep dive is judgment reps(coming Jul 13), and the instrument is the decision log(coming Jun 26).

Self-test: can you produce your last ten significant decisions, with the evidence present at the time and what happened after? If the answer lives in nobody's file, your org learns by anecdote.

Acquire it: keep a decision log for one quarter. Read Annie Duke, Thinking in Bets. Run one premortem (Gary Klein's prospective hindsight method) on your next big bet.

3. Taste

The trained eye for what is good before the metrics confirm it. When everyone can generate ten plausible versions of anything in an hour, choosing well among abundance becomes the differentiator. I argued in taste is the last moat that this is now the scarcest input in product work.

Self-test: put three AI-generated versions of your next feature copy or flow side by side and rank them with written reasons. Then check whether a designer or your best customer ranks them the same way.

Acquire it: volume plus feedback. Critique sessions with designers, building your own 60-minute prototypes until the difference between fine and good is felt, and studying products you admire one screen at a time.

Layer 2: Expression

Judgment that cannot be transmitted does not exist, organizationally speaking.

4. Storytelling

Not presentations. The structuring of evidence into an argument that survives retelling when you are not in the room. SCQA, the Minto pyramid, working backwards from the press release. This is the single most undertrained skill in product management and the full deep dive is storytelling is a PM core skill(coming Jul 6).

Self-test: hand your last strategy memo to someone outside your team. An hour later, ask them to repeat the argument. If what comes back is the topic but not the argument, you wrote a report, not a story.

Acquire it: Barbara Minto, The Pyramid Principle (the book; the consulting world runs on it). Wes Kao's Executive Communication course on Maven and her frameworks on Lenny's podcast. Bryar and Carr, Working Backwards, for the PR/FAQ discipline, then write one with the template(coming Jul 1). On this site: show, don't tell and the board narrative chapter.

5. Writing for machines

Half your readers are now agents: the agents that draft your team's code, the AI overviews that decide whether customers find you, the fleet that runs your own workflows. Writing specs, evals, and prompts that machines execute faithfully is a distinct craft from writing for humans, with its own failure modes. Deep dive in writing for machines(coming Jul 16).

Self-test: take your last PRD or prompt and run it through a model cold, no verbal context. Does what comes back match your intent? The gap is your score.

Acquire it: Prompt Ops on this site, Anthropic's prompt engineering documentation, and the discipline of the eval as spec, which forces precision no prose spec ever forced.

6. Prototyping

The argument you can touch. A working prototype ends debates that decks prolong, and it is now a two-hour skill, not an engineering request. Prototype Before You Spec is the foundational chapter.

Self-test: could you put a working prototype of your current top idea in front of a customer this week, yourself, without filing a ticket?

Acquire it: install Claude Code or equivalent and ship one small real thing this week, per the Builder PM Week 1. Then use the mob prototyping pattern to spread it to your team.

Layer 3: Systems

The layer where PM work became infrastructure.

7. Evals

Defining quality as labeled examples and measurable criteria, then holding the product to them continuously. For anything AI-touched, the eval is the spec, the acceptance test, and the early-warning system in one artifact. This is currently the rarest skill on the stack and the fastest route to being indispensable.

Self-test: what is the eval score on your most important AI feature, and which way is it trending? If you cannot answer today, neither can your org.

Acquire it: The Eval Is the Spec, then build the five-row eval this week. Outside: Hamel Husain and Shreya Shankar's AI Evals course and Anthropic's evaluation guide.

8. Agent orchestration

Designing, briefing, and quality-controlling a fleet of agents that do the mechanical layer of your job. The skill is not prompting, it is delegation architecture: what to hand off, what to keep, and how to verify. The editor's stance is covered in PM as Editor and the org-level view in PM as a Team of AI Agents.

Self-test: how many hours of your last week were mechanical work an agent could have done? If you do not know the number, the answer is too many.

Acquire it: stand up three agents from the fleet, run them for a month, and tune them with the agent tuning playbook. Outside: Anthropic's agent-building guides and Simon Willison's weblog for the practitioner's running commentary.

9. Signal architecture

Designing where truth enters your building: which customer signals get captured, synthesized, and routed to whom, continuously and mostly automatically. Discovery stops being a ritual and becomes plumbing. The foundation is Continuous Discovery on Autopilot and Continuous Listening.

Self-test: when a customer told your company something important last Tuesday, what happened to it? Trace one real signal end to end. The number of manual hops is your architecture's debt.

Acquire it: build the call triage pipeline first, then the discovery agent stack.

Layer 4: Economics

The layer that got promoted from finance's problem to yours.

10. Unit economics

Knowing what your product costs to run per outcome, and what moves that number. AI products carry real marginal cost, which means PMs now own a P&L line whether anyone told them or not. Gross Margin Is Your Job Now is the chapter; financial fluency for PMs(coming Jul 9) is the skill-building deep dive.

Self-test: what is the gross margin on your product or feature area, and what are its two biggest drivers? If your answer is "that's finance's number," that is the gap.

Acquire it: the deep dive above, then Berman and Knight, Financial Intelligence (the standard text for non-finance managers), then one working session with your finance partner where you rebuild your product's cost per outcome together. Practice the conversation in the CFO margin scenario.

11. Pricing

The highest-leverage decision most PMs never touch. AI broke per-seat logic, and pricing migrations are now product work, not pricing-committee work. Start with Pricing for AI Products and the 18-month migration playbook.

Self-test: can you articulate why your product is priced the way it is, what value metric it should be priced on, and what evidence supports the gap between the two?

Acquire it: the chapters above, plus April Dunford's Obviously Awesome for the positioning layer underneath pricing, and Madhavan Ramanujam's Monetizing Innovation for the willingness-to-pay discipline.

Layer 5: Leadership

The layer that decides whether any of the above scales past you.

12. Coalition building and killing things well

Two skills that are one skill: moving an org. Evidence does not implement itself, and nothing tests a leader like retiring something people are attached to. The coalition mechanics are in the coalition map essay; the retirement craft is in The Deprecation Playbook and The Anti-Backlog.

Self-test: name the last thing you killed publicly, with a written explanation, that someone senior wanted kept. If nothing comes to mind, you have been adding, not leading.

Acquire it: there is no course. Run the CPO 30/60/90 coalition map at whatever scale you operate, kill one zombie initiative this quarter using the playbook, and study the org-shape essays: the new org chart and hiring the Builder PM.

How to use this chapter

Run all twelve self-tests in one sitting. Score yourself honestly: have it, building it, don't have it. Then spend the next two quarters on your two lowest scores, not your two favorites. The favorites are favorites because you already have them.

For IC PMs: evals and storytelling first, almost always. For new CPOs: the CPO 30/60/90 will surface your gaps faster than introspection will; most arrive missing economics or the killing skill.

One warning. The stack is not a curriculum to complete, it is a portfolio to rebalance forever. The 2021 stack looked permanent too.

Pick one thing this week

Run the twelve self-tests tonight. Write the three-column score in your notes. Pick the worst one and book two hours this week on its acquisition path. That is the whole move. The compounding does the rest.

Sources: Barbara Minto, Wes Kao on Maven, Teresa Torres, Product Talk, Bryar & Carr, Working Backwards, Annie Duke, Thinking in Bets, Gary Klein on premortems, HBR, Hamel Husain & Shreya Shankar, AI Evals, Anthropic documentation, Simon Willison, April Dunford.