Product Management for Engineers · Page 5

Who Decides and What Carries the Decision: Product Roles and Spec Artifacts

Teams differ less in who has a title than in who holds each decision on the way from "which problem" to "did it work", and in which document carries a decision across the hand-off to the people who build. This page lays out six role models, the Scrum product owner as a seventh reference point, the engineer's part in each, the AI-era shifts in the PM job, and five spec artifacts with their author, cost and the risk each reduces. Afterwards you can say which decisions you own or can claim under a given team model, what the PM is accountable for in it, and which artifact fits a given decision.

Deep dive 60 min read 17 sources As of 2026-10-26
Eight product decisions from "which problem" to "did it work", and which of five spec artifacts carries each decision according to its source Decision PRD Prototype Pitch PR/FAQ Project spec 1 Which problem 2 Target and measure 3 Solution concept 4 Scope cuts, appetite 5 Priority and capacity 6 Tasks, implementation 7 Ready to ship 8 Did it work yes yes yes yes part yes part yes yes yes yes yes part yes part yes part yes yes yes part the source says the artifact carries it partly or by implication blank: not carried
Figure 1. The decision ladder this page uses, upstream at the top. Each cell is the writer's reading of what the artifact's own source says it contains (synthesis, not a source claim): PRD1, prototype as spec2, pitch3, PR/FAQ4, project spec56. Two things to see: only the PRD's release criteria say when to ship, and no artifact carries rung 6 or the full answer to rung 8. Every artifact stops at the hand-off; what happens after it is decided by whoever the team model gives that decision to.

1. Decisions, responsibility and authority

1.1 Eight decisions on one ladder

Job titles do not say who decides. A team that calls someone a product manager can give that person every rung of the ladder in Figure 1, one rung, or none. The ladder is the writer's construction synthesis, chosen so that every source on this page can be placed on it. Horowitz's memo compresses it into "what" and "how"; Scrum into ordering and a plan; Shape Up into shaping, betting and building. Every model below is a position of that what/how line plus a mechanism for the people on each side of it.

A spec artifact is the thing that crosses the line. That is why the two halves of this page, roles and artifacts, belong together: a PRD that states the problem and leaves the solution to engineers is a statement about who decides the solution; a pitch that sketches the solution is a different statement. Sections 2 to 9 cover the role models and the engineer's place in them. Section 10 covers the artifacts.

1.2 Responsible, accountable and in charge are three different claims

Cagan assigns each of the four risks to one role. The product manager is "responsible for the value and viability risks, and overall accountable for the product's outcomes"; the designer is responsible for usability and "overall accountable for the product's experience"; the lead engineer is responsible for feasibility and "overall accountable for the product's delivery"7. The essay says nothing about who may overrule whom. It is a map of who answers for what.

Eriksson's argument against the "CEO of the product" label is about the other column. "Product managers simply don't have any direct authority over most of the things needed to make their products successful", and "even today's most senior product leaders only have hiring and firing control over their direct reports"8. He does not dispute accountability: "most product managers don't pass the buck, and this ownership of the problem and willingness to make the hard decisions is where a product manager gets closest to being a CEO", and a PM "should live and die by your product's success and failure"8. The Scrum Guide makes the opposite structural choice. The product owner is "accountable for maximizing the value of the product", is "one person, not a committee", and "for Product Owners to succeed, the entire organization must respect their decisions"9. Horowitz's analogy rests on responsibility ("takes full responsibility and measures themselves in terms of the success of the product") plus management of the team: "they are not part of the product team; they manage the product team"10.

Read together, the sources agree that someone is accountable for outcomes. They differ on whether that person may command, and on whether accountability implies a veto over the people who design and build. Table 2 and Figure 4 keep the two columns apart for that reason.

1.3 What kind of evidence this is

Every model on this page is described by the organization or author that practises it. No fetched source compares PM-led with engineer-led teams on any outcome, and the nearest thing to a test is one company reversing itself, reported by its own co-founder (Section 4). The page therefore treats each model as a mechanism with a stated failure mode it answers and a stated cost it accepts, not as a result. Three kinds of statement are kept apart: cited statements from a source, arithmetic the writer derived (labeled derived), and readings across several sources (marked synthesis). Source ages differ by decades, which matters for the first model, so the chronology is collected at the end of the page.

2. The PM as CEO of the product

2.1 What the memo assigns, and the failure it answers

Horowitz's memo states the model in its first sentences: "A good product manager is the CEO of the product." The line it draws is between target and method: "Good product managers crisply define the target, the 'what' (as opposed to the how) and manage the delivery of the 'what.'" The inverse is a diagnosis: "Bad product managers feel best about themselves when they figure out 'how'"10. The PM is also a manager: "They are not part of the product team; they manage the product team", as "the marketing counterpart of the engineering manager". The medium is writing: "Good product managers communicate crisply to engineering in writing as well as verbally", "don't give direction informally", and "take written positions on important issues (competitive silver bullets, tough architectural choices, tough product decisions, markets to attack or yield)"10. The PM's job is to point the team at "revenue and customers"10.

The model exists because of a specific failure, and the memo names it: bad product managers "define good products that can't be executed or let engineering build whatever they want (i.e. solve the hardest problem)"10. Cagan's 2005 PRD paper, which cites Horowitz's notes as a partial source, restates it and gives the mechanism. The danger is "the engineering organization overwhelming the product manager with requirements and essentially defining the product they are to build, either based on a technology they are excited about, or because they are challenged by a particularly hard problem they want to solve", and the paper closes the point with the memo's own phrase, "let engineering build whatever they want (e.g. solve the hardest problem)" (p. 17)1. For an engineer who builds inference runtimes this is worth reading as a description of a real pull, not an insult. When nobody else owns the question "does this move revenue or customers", the hardest technical problem is the default choice, and the CEO model is a rule against that default.

2.2 A dated memo, and one lineage rather than two sources

The memo's own host attaches a warning: "This document was written 15 years ago and is probably not relevant for today's product managers. I present it here merely as an example of a useful training document"10. The page is dated June 15, 2012, so the memo dates from about 1997 (derived: 2012 minus 15); Eriksson, writing in 2017, calls it "nearly 20 years ago"8. It invokes Barksdale as the model executive; Cagan's paper identifies him as the former CEO of Netscape1. The body names no PRD, discovery, prioritization, roadmap, metric or prototype; the only goals it states are revenue and market share10.

It is also not independent of the 2005 PRD paper. Cagan's footnote 2 says the paper "is based in part on the notes and insights of Ben Horowitz"1. Agreement between those two documents is one lineage, and the same author later reversed the artifact emphasis (Section 10.3). Treat the CEO model as a dated position that still frames the debate, not as an accumulated finding.

2.3 The formal cousin: Scrum's product owner

The Scrum Guide does not use the term product manager, so this is a comparison of decision statements, not of jobs. Its product owner holds the "what" and the order: developing and communicating the Product Goal, "creating and clearly communicating Product Backlog items", "ordering Product Backlog items", and being "one person, not a committee". Others "wanting to change the Product Backlog can do so by trying to convince the Product Owner", and "only the Product Owner has the authority to cancel the Sprint"9. The Developers hold the "how" and the amount: "How this is done is at the sole discretion of the Developers. No one else tells them how to turn Product Backlog items into Increments of value", and "the Developers who will be doing the work are responsible for the sizing"9. The team is "self-managing, meaning they internally decide who does what, when, and how", and "within a Scrum Team, there are no sub-teams or hierarchies"9.

That is Horowitz's line with one change. The authority is over the backlog, not over people: the product owner decides what the team works on and in what order, and does not manage the Developers. The Guide also states a condition on the organization, "the entire organization must respect their decisions"9, which is the part Eriksson says a PM cannot count on having. The Guide presents no evidence that the arrangement works; it is a definition published by the framework's authors, dated November 20209.

2.4 What the engineer gets and gives up

Under both variants the engineer's "how" is explicitly protected: by implication in the memo, by rule in Scrum. In exchange the "what" belongs to someone else, and the engineer's way to affect it is to supply information and persuade. The cost the critics name is on the PM side. Eriksson reports that PMs who adopt the CEO label "tend to believe they have all the answers, that they produce the best solutions and designs, and that their teams should just do what they're told", with "disastrous results"8; his evidence is what he has seen, and the page contains no data. The next section covers the answer he and Cagan give.

3. No authority: team sport, the trio and the surgical team

3.1 Authority is earned, and the PM answers anyway

Eriksson's position is that a PM is not the CEO of anything unless he is also the founder. His list of what a PM lacks is concrete: direct authority over "user and data research through design and development to marketing, sales, and support"8. What a PM does instead, in his account, is lead "through communication, vision, and influence". He quotes Ken Norton: "When you're a product manager, you're generally not the boss. You need to gain authority through your actions and your leadership skills, not your role"8. The metaphor he prefers is a team in which "the best teams don't have bosses - they have coaches"8. The failure mode this answers is the one from Section 2.4, the "mini-CEO" who tells the team what to do. The cost it accepts is that the PM is accountable for outcomes without the means to compel anyone, so the job depends on persuasion and on the team's willingness to be persuaded.

3.2 Deference by domain: the surgical team

Cagan's April 2026 FAQ turns the same stance into a decision rule. "No, the PM is not 'the decider,' and it's quite dangerous to think of yourself in this light." Each member of the cross-functional team, designers and engineers included, "uses their experience and judgement to make dozens or even hundreds of decisions every day." The model he offers is a surgical team: "Rather than having 'a decider,' the surgical team defers to the person that is best suited for the particular decision. If there are impacts on other areas, we collaborate to determine the best solution"2. The PM is "an individual contributor, just like the product designer and the engineers", and "you are a builder, not a manager"2. The PM's specific contribution is that "you are responsible for the value and viability of the proposed solutions", which is the allocation in Cagan's 2017 risk essay7 restated as a duty to the team.

The FAQ also contains the one explicit division of decisions between team and leadership that any source in this set gives. "Your product strategy is what picks the problem, and this is usually done by your product leaders." The team's side is the solution: "You are trusting that your product leaders and stakeholders will identify worthy problems to solve, and they are trusting that you can solve these problems in ways that work for the customer and the business"2. In the ladder of Figure 1, rung 1 belongs to leaders and the solution, from rung 3 on, to the team; the FAQ is silent on who sets the target (rung 2) and the capacity (rung 5). The trio itself is Torres's definition, "a product manager, a designer, and a software engineer", set up as a team activity11. The 2026 FAQ names no tie-break between members. "Defers to the person best suited" is the rule, and who judges "best suited" is not stated.

3.3 Why a single decider cannot be the mechanism (derived)

The "dozens or even hundreds of decisions every day" figure is per team member2. Take it at face value and ask how much of that a single decider could personally take. Let \(n\) be the team size, \(d\) the decisions per person per day, \(f\) the fraction routed to one decider, and \(m\) the minutes the decider spends on each. Then the decider's daily load and the largest fraction that fits in an \(H\)-hour day are

$$ L = \frac{n\,d\,f\,m}{60}\ \text{hours}, \qquad f^{*} = \frac{60\,H}{n\,d\,m}. $$

Illustrative inputs, all assumed except the range of \(d\): \(n=6\), \(m=2\) minutes, \(H=8\), and \(d\) from 24 (the low end of "dozens", reading a dozen as 12) to 100 (the low end of "hundreds"). Routing everything through one person at \(d=24\) means \(6 \times 24 = 144\) decisions, \(144 \times 2 = 288\) minutes, 4.8 hours: 60% of the day on approvals alone. At \(d=100\) it is 600 decisions and 20 hours, more than a day. With these inputs \(f^{*} = 480/(12\,d) = 40/d\), so the decider saturates at 83% of decisions for \(d=48\) and at 40% for \(d=100\) (Figure 2). The number is not a prediction about any team. It shows that if the sourced decision rate is even roughly right, most decisions must be made without the PM, and that the only question is by whom. Page 3 (Section 5) shows what waiting does to a queue as its server approaches full utilization; a single decider is such a server.

Hours per day a single decider needs, against the share of a six-person team's decisions routed through them, for 24, 48 and 100 decisions per person per day; the 8-hour line is crossed at 83% and 40% for the upper two 0 4 8 12 16 20 0% 25% 50% 75% 100% share of the team's decisions routed through one decider decider hours per day 8-hour day d = 24 d = 48 d = 100 83% 40%
Figure 2. Derived load on a single decider, \(L = n\,d\,f\,m/60\), with \(n = 6\) people and \(m = 2\) minutes per decision (both assumed for illustration) and \(d\) decisions per person per day. Only the order of magnitude of \(d\), "dozens or even hundreds of decisions every day" per team member, is from the source2; the lines and the two crossing points (\(f^{*} = 40/d\)) are the writer's arithmetic.

Single-decider load (derived; only the range of decisions per person is sourced)

Decisions routed per day
Decider hours per day
Share of an 8-hour day
Largest share one person can take

Formula: \( L = n d f m / 60 \) hours; \( f^{*} = 480 / (n d m) \) for an 8-hour day, capped at 100%. Defaults reproduce the 144 decisions and 4.8 hours in the text.

4. Engineer-led at PostHog

4.1 The stated model and the failure it answers

PostHog's co-founder and a colleague open with the failure they are reacting to: "Too often product managers exist to control engineers, or shield them from organizational dysfunction." In that arrangement "PMs become the bottleneck and gatekeeper for all decisions", and engineers "get a sanitized version of the truth, so they don't have the right information to make the best decisions"12. Their answer moves the decisions: "At PostHog, PMs no longer own the roadmap, make product decisions, or shield engineers from users or the wider business goals. Instead, our engineers run the show. They manage product teams"12. The handbook agrees on the operating level: "Engineers own sprint planning and spec'ing out solutions"13. A second handbook page names the nearest thing to a decider: "The closest thing we have to a 'product owner' at PostHog is the team lead", and "engineers choose what to work on next based on shared context"14. The PM can challenge but does not overrule: "PMs can and should challenge the team's decisions, but ultimately it's the engineers who make the final call"12.

The founding argument is one story. An engineer saw customers asking for session replay and wanted to build it; the CEO "thought it was a terrible idea"; the engineer "disagreed and built it anyway", and it "ended up being wildly popular with customers, and changed our entire company strategy". The conclusion drawn is general: "The reason Karl was right – and why engineers are usually right – is they have the deepest understanding of what can be built"12. That is a CEO's causal claim from a single success; the post gives no count of engineer-originated ideas that failed.

4.2 What the PM holds: domains, not the roadmap

The handbook page on the PM role is more specific than the essay. It splits the PM's work into three tiers (Table 1). The essay says PMs no longer "make product decisions"12; the handbook says the PM owns three areas, one of them pricing14. Both can hold only if "product decisions" means the roadmap and what to build, not every decision. The page adds a mechanism for the contested middle: for "major calls (like positioning, pricing, or launch scope)", the PM and team lead "select an informed captain: One person who makes the decisions and sees them through. The other can provide feedback when requested, but focuses elsewhere"14. It does not say who is captain in which case. synthesis This is deference by domain, the same rule as the surgical team in Section 3.2, written down as a policy with a named tie-breaker for the cases that cross domains.

Table 1. The PM role at PostHog as the handbook's PM-role page states it, 2026-10-26 snapshot14. "Owns" is the page's own verb; the page says "shares ownership" and "supports" for the other two tiers.
TierAreas, in the page's wordsNote
OwnsSystems for user discovery and product thinking; team metrics ("always know how the product is performing, both in usage and revenue"); pricing ("regularly review whether pricing aligns with the value users get")Includes the monthly growth review13; pricing rules are page 6's subject
Shares with the teamUser understanding; monitoring the competitive landscape and industry trends"Product engineers are equally expected to contribute"
SupportsLaunching a product in beta or general availability; "deciding on what we should be building next"; feedback on UXThe PM "might own projects" here "in consultation with the team"
Does not own"roadmaps"; does not "dictate what to build next"Team lead is the closest to "product owner"

4.3 The PM arrives with the revenue

The same page says when a PM joins. "Especially in the early stages (1 & 2), a team usually doesn't have a PM yet — product engineers own all product decisions", and at those stages "it's typically the team lead's job to decide: what to build, when to release the first version, how to charge for it". In most cases "we add a PM after a product has launched and started generating revenue (stage 3)"14. The product-team page lists one team with no PM and does not say which stage it is at13. There is one stated exception, and it matters for an inference-runtime engineer: "When a team is working on an infrastructure-heavy product, it's more important that the team lead has strong engineering and infra skills than product skills. In that case, it can make sense to add a PM early who can focus on product direction, positioning, and user research while the lead focuses on technical execution"14. A runtime team plausibly sits inside that exception, in which case PostHog's own policy would put a PM beside the technical lead early rather than late.

PostHog's four product stages, who decides at each, when a PM joins, and the exception for infrastructure-heavy products 1 Idea to beta 2 Beta to GA 3 GA to growth 4 High revenue first version inusers' hands a product userswant to pay for strengthen andexpand adoption sustain growth,manage complexity Who decides Team lead: what, when, price Engineers choose; PM gives context PM on the team usually none yet added after launch and revenue exception: infrastructure-heavy, PM early PM's question Who is it for? How do we charge? Are users getting value? What limits growth?
Figure 3. PostHog's stage rule as its handbook states it: who decides, when a PM is added, and the infrastructure-heavy exception. Stage goals and the PM's key questions are the page's own table; stages 1 and 2 are where "product engineers own all product decisions"14. One company's policy as of 2026-10-26, not a general finding.

4.4 The conditions the sources attach to the model

The sources give no outcome comparison, but they state what the model leans on. synthesis Each item below is a source statement; the grouping into "fences" is the writer's.

  • Trust. "Of course, this requires an extremely high-level of trust in your engineer's decision-making"; "a giant checklist of processes and reviews is a tell-tale sign that a company doesn't trust their engineers"12.
  • Small teams with an engineer in charge. "No more than six people per team"; "Split the team when it hits 6 people. No exceptions"; "Put an engineer in charge"12. Section 3.3 used six as its illustrative team size.
  • Demand someone else has shown. Before product-market fit, "we only build things that already have successful competitors with real revenue", so "just build it" works "disproportionately well" because the value question has been answered elsewhere; "just shipping is often the best way to validate an idea"15. The same page recommends staggered rollouts and descoping to limit what a wrong guess costs15.
  • A feedback loop with a decider on each side. The PM compiles a monthly growth review; "the product team lead (an engineer)" decides whether to continue or change course; "the CEO or senior leader" challenges assumptions and holds the team accountable. "Without this accountability, autonomy can become directionless"12. How the review works is page 4's subject (Section 10.1 there); here the point is the split of rights: compile, decide, challenge.

The handbook's own list of "bad product engineers" shows what the trust is spent on: they "only work on things they've been told to work on", "never challenge why they're being told to work on something", and "don't track if the things they've built are being used"15. If those fences are absent in your organization, you cannot import PostHog's decision rights by asking for them; the rights were granted against specific conditions.

4.5 The reversal, the counts, and an open gap

The essay's opening is the nearest thing to a test. "When Tim and I first started PostHog in 2020, I was adamant we would never hire a product manager", and four years on, "I admit I was (partially) wrong. We need product managers. In fact, we couldn't have shipped 8+ products, or hit our revenue goals without them"12. The same document that credits an engineer with a strategy change credits PMs with scale, and gives no figures for either. The count "8+ products" is the essay's; the handbook confirms the operating model, not that count13.

The handbook's product-team page lets you count what the model looks like in practice. As fetched on 2026-10-26 it lists seven PMs, each attached to between two and four teams (21 team entries in all), and one team marked as having no PM (derived from the page's lists)13. The same page's Q2 2026 theme text refers to "a group of 5 (soon to be 7)", so the page carries two PM counts and is a living document13. Two derived bounds follow. Under a six-person cap, a PM attached to two to four teams touches at most 12 to 24 team members; the cap is a limit, not an average. And the page's rule of thumb, "1 interview per week per PM", spread across two to four teams, is one PM-run interview per team every two to four weeks13; engineers are expected to talk to users as well, and the engineering handbook says they can book their own interviews15. One more item belongs here. Among the PM team's Q2 2026 themes is to "think about ideas how to fix the 'accountability gap' in growth reviews": more accountability expected from product teams, and growth reviews done early in the month by PMs13. The accountability half of the model is an open problem at the company that wrote it.

5. Engineer-led at Linear, with one PM

5.1 Rotating project leads and a short spec

Linear's account is given by its own product and engineering staff. One engineer says it plainly: "We're still fairly small and only have one PM (Nan, our Head of Product), so we have always had engineers leading projects", adding that at the larger companies where that engineer worked before "it was always the PMs leading everything", and that at Linear "we get to influence the product roadmap quite a bit"6. The leads rotate: "The project leads are rotational. Nobody is the de facto project lead." The lead "drives the communication around the project, shares updates, and ensures things are running on time", which "allows the other engineers to focus on coding"; relevant past expertise and available bandwidth are weighed, and "everybody has to take a turn and do what nobody wants"6. A project is defined by size: "If multiple people work on it, and it takes over two weeks, that's a project"6. Linear's published method adds the ownership rule: "Every project should have a named owner who's responsible for writing the project brief and delivery", and "even if other people are involved, the responsibility should lie with a single person"5.

The decisions the account places with the project team are rungs 3 and 4 of Figure 1. The roadmap is "a high-level roadmap with themes that we want to address in the next six months or so", broken into features that "often become the projects themselves". The descriptions are "quite open-ended to start with": about a year before the post, "we knew we wanted to build an analytics feature", but "it was up to the project team to define what that actually meant", and "what might have looked like a single project at the beginning turned into four or five different projects in the end"6. Readiness is decided in the team: "There's no strict process or a document with checkmarks to tell you when the project is ready for release. It comes down to the engineers working on the project feeling that the project is ready"6. Debate runs on product questions: "Most of our discussions and debates are around product decisions rather than technical implementation"6.

5.2 What the account does not say

The post does not say who has final authority over what gets built or how projects are prioritized, and it does not describe the PM's decision rights; the PM appears collecting inputs ("feedback, backlogs, user interviews, and random tickets") and synthesizing "the common themes and open questions"6. The method page never mentions product managers5. It is also a vendor account: Linear sells project-planning software and is describing its own planning practice. Its date is October 5, 2023, so the single-PM statement is three years old at this page's as-of date6.

The model has a stated condition. "Fewer dedicated PMs are required when you are still small, structure your teams properly, and have a fairly senior team"6, and another engineer credits the team being "quite senior" for efficient engineering discussion6. That is a statement about size and seniority, so it sits alongside PostHog's reversal (Section 4.5), not against it: PostHog added PMs as it added products. synthesis The two engineer-led models also differ in one mechanism. PostHog fixes an engineer as team lead and says "put an engineer in charge"12; Linear treats leadership as a duty every engineer takes in turn, to spread the skill and "build empathy within the team for others"6. Rotation lowers the cost of one lead leaving or burning out and raises the cost of a lead's learning curve; the post states the first benefit and no cost.

6. Shape Up: a betting table with no PM seat

6.1 Three hand-offs: shapers, table, builders

Shape Up gives decision rights to three groups in sequence, and a PM is in none of them. Shapers produce the concept. "Shaping is creative and integrative. It requires combining interface ideas with technical possibilities with business priorities", which takes "a generalist" or "one or two other people". The shaper "need[s] to be technically literate" but not a programmer: "You should be able to judge what's possible, what's easy and what's hard"3. The work is strategic as well: the shaper asks "what counts as success?" and "which customers are affected?"3. It is also private: shaping-track work "is kept private and not shared with the wider team until the commitment has been made to bet on it", which lets shapers drop a concept without anyone having planned around it3.

The betting table decides what is built next and by whom. At Basecamp it "consists of the CEO (who in our case is the last word on product), CTO, a senior programmer, and a product strategist"3. The author calls the arrangement final: "The highest people in the company are there. There's no 'step two' to validate the plan or get approval. And nobody else can jump in afterward to interfere or interrupt the scheduled work", and the call "rarely goes longer than an hour or two"3. The decision criteria are page 3's subject; the role fact is that two of the four seats are technical, the CTO's and a senior programmer's. The build team, one designer with one or two programmers plus QA3, gets the project and not the tasks: "Nobody plays the role of the 'taskmaster' or the 'architect' who splits the project up into pieces for other people to execute." The team "is going to define their own tasks and their own approach to the work", with "full autonomy", inside the boundary the pitch drew: "we aren't giving the teams absolute freedom to invent a solution from scratch. We've done the shaping. We've set the boundaries"3. Deploying within the cycle ("done means deployed") is the team's job3.

6.2 The failure it answers, and what the text leaves open

The text names two failures. One is work handed over at the wrong level of detail: "Product managers often err on one of these two extremes", wireframes that leave designers no room or a few words that leave programmers guessing3. The other is splitting a project into tasks up front, which is "like putting the pitch through a paper shredder", and treats "talented people" like "'code monkeys' or ticket takers"3. So product managers appear in the fetched chapters twice outside any role in the process: in the introduction's list of people with a problem ("Product managers can't find time to think strategically about the product") and in Chapter 2 as the example of over- and under-specification3. That is the text's own word choice. It does not say Basecamp has no PM, and the fetched pages assign no task in shaping, betting or building to one.

Three things the chapters leave out matter for an engineer who wants a seat. Chapter 6 does not say who writes a pitch3. No chapter fetched says what counts as success for a bet after the cycle, which is rung 8 of Figure 1. And the structure depends on one condition the author states outright: "buy-in from the very top is essential to making the cycles turn properly"3. The table works because the people with the final word sit at it; a team that shapes well for a table that does not include its decision-makers has copied the form without the condition. synthesis

7. The models side by side

7.1 Each model answers a failure, and the failures are mirror images

Table 2 lines up what each source says about the failure it is built against, who has the final call on what is built, what the PM is for, and who answers for outcomes. synthesis The failure column is the useful one. The CEO model and the Scrum arrangement guard against engineers choosing the hardest or most interesting problem; the engineer-led models guard against the PM as a gate on every decision; the "not the CEO" position guards against the PM as commander; Shape Up guards against work handed over at the wrong level of detail. Each model removes one failure by moving decision rights, and each moves them toward the party the next model distrusts. The sources describe the cost of the failure they fight; of the costs of their own remedy they state only conditions: trust and small teams at PostHog, size and seniority at Linear, buy-in from the top at Basecamp.

Table 2. Six models and the Scrum product owner: failure each answers, final call, PM's stated mandate, and who answers for outcomes. Every cell is the source's own claim about its own practice; no source reports outcome data.
Model (source, date)Failure it answersFinal call on what is builtPM's stated mandateAnswers for outcomes
CEO of the product (Horowitz memo, about 1997, posted 2012)10Engineering building "whatever they want"PM defines the "what"Manages the product team; direction in writing; revenue and customersPM: "full responsibility"
Scrum product owner (2020)9Not stated in the GuidePO orders the Product Backlog; Developers decide the "how"None by name; the Guide does not use "product manager"PO "accountable for maximizing the value"; Developers for a usable Increment
No authority: trio, surgical team (Eriksson 2017; Cagan 2017 and 2026)872The "mini-CEO" who commands; viability under-estimatedLeaders pick the problem; the team "defers to the person best suited"Responsible for value and viability; builds and tests prototypesPM, "overall accountable for the product's outcomes"
Engineer-led, PostHog (essay 2024, handbook 2026)1214PM as "bottleneck and gatekeeper"Engineers; team lead is the nearest "product owner"Owns discovery systems, team metrics, pricing; "compass"Team lead decides after the PM's review; CEO challenges
Engineer-led, Linear (2023)65Not statedNot statedOne PM; inputs and synthesis described, rights not statedNamed project owner, a rotating engineer
Shape Up betting table (web edition)3Tasks handed down; work at the wrong level of detailBetting table: CEO ("last word on product"), CTO, senior programmer, product strategistNone named in the processNot stated for results

7.2 The matrix: where the lines fall

Figure 4 puts the same sources on the ladder of Figure 1, with each cell colored by who the source says holds the call. Three patterns stand out, all readings of the writer's own cells. First, rungs 6 and 7, tasks and readiness, belong to builders in every model that speaks to them, including the CEO model, whose "how" is engineering's. Second, rung 1 is the contested one: the PM or product owner in two columns, leadership in two (the trio's strategy and Shape Up's table), engineers in one. Third, rung 8, whether it worked, is the emptiest row: two columns are silent, and only PostHog describes a mechanism with named roles (compile, decide, challenge). Page 4 takes up how outcomes get reviewed.

Decision-rights matrix: who holds seven decisions under the CEO model, Scrum, the trio, PostHog, Linear and Shape Up, according to each source; dashed cells are decisions the source does not address CEO model Scrum Trio PostHog Linear Shape Up 1 What to build next PM PO Leaders Eng lead — Table 3 Solution concept PM / Eng Dev Trio Eng Team Shapers 4 Scope and sizing — Dev — Eng Team Shapers 6 Tasks, implementation Eng Dev Eng Eng Eng Build team 7 Ready to ship — Dev — Eng Eng Build team 8 Did it work PM PO PM Lead+exec — — Writes the spec PM PO PM Eng Lead Shapers PM or product owner engineers, builders leadership, table shared or another role source silent
Figure 4. Decision-rights matrix: who holds each decision under each model. Row numbers are the rungs of Figure 1. Color is the party the source names; dashed cells are decisions the fetched source does not address. Cells are the writer's reading of Table 2 and Sections 2 to 6 synthesis; sources: CEO memo10, Scrum Guide9, Cagan's risk essay and FAQ72, PostHog12131415, Linear65, Shape Up3. Eriksson's position8 allocates no decisions, so it has no column; it appears in Figure 5.

7.3 One axis, ordinal, and two models that do not fit it

Figure 5 orders the sources that state how much authority the PM has over what is built. The positions are ordinal, set from each source's own sentence, and the gaps between markers carry no scale. synthesis Linear and Shape Up are drawn off the axis: their sources do not state PM authority at all. They are the two models whose sources say most about builder autonomy in rungs 4 and 5.

Ordinal spectrum of how much decision authority each source gives the PM over what is built, from manages the product team to engineers make the final call; Linear and Shape Up are off the axis because their sources do not state it PM holds the call engineers hold the call CEO model manages the product team Scrum PO organization must respect decisions Trio responsible for value, viability Not the boss no direct authority PostHog engineers make the final call Off the axis: the source does not state PM authority Linearengineers lead; PM rights not stated Shape Upno PM seat at the betting table
Figure 5. Ordinal spectrum of the PM's authority over what is built, one marker per source position: Horowitz10, the Scrum Guide9, Cagan's risk essay7, "not the boss" from Eriksson and the 2026 FAQ82, and PostHog12. Positions are the writer's reading of each source's own statement; distances are not measured.

7.4 How many independent confirmations there are

Table 3. Relations among the sources on roles, and what they imply for how much weight agreement carries.
SourcesRelationConsequence for the evidence
Horowitz memo and Cagan's 2005 PRD paperThe paper is "based in part on the notes and insights of Ben Horowitz"1One lineage; agreement is not confirmation
Cagan 2005 (PRD), 2017 (four risks), 2026 (FAQ)One author over 21 years172One voice changing; read as before and after, not three confirmations
PostHog essay and three handbook pagesOne company; the essay is co-written by its CEO12Counts such as "8+ products" and the split at six are self-report; the handbook confirms the operating model, not the counts
Linear Method and "How we run projects"One company, which sells planning software56Practice as described by those who run it
Shape UpOne company's published method3Experience and anecdote, no measured data
Eriksson, Orosz, NgSingle authors; Eriksson quotes Ken Norton and Uber's first Head of Product, Orosz quotes an Atlassian PM and a Shopify engineering head81617Opinion; secondhand quotes add voices, not data
Scrum GuideIts authors define the framework9A definition, not evidence that the arrangement works

Counting organizations rather than documents, three describe an engineer-led or PM-light model (PostHog, Linear, Basecamp), each describing itself, and none compares its results with a PM-led alternative. The PM-led side is a 1990s memo, its 2005 derivative, the Scrum definition and Cagan's later essays. Where the page says a model "works", the claim means "its practitioners report it works".

8. The product-minded engineer

8.1 Nine traits, read as risks

Orosz defines the type by interest: "Product-minded engineers are developers with lots of interest in the product itself. They want to understand why decisions are made, how people use the product, and love to be involved in making product decisions"16. His list has nine traits, each from observation of colleagues; the page cites no data16. Figure 6 maps them onto the four risks from page 1. synthesis The mapping is the writer's: a filled mark means the trait's text names the risk, an open mark means it is inferred. Of the 16 marks, 12 are filled. The count that matters is by column. Seven of nine traits touch value risk. Feasibility, the risk an engineer already owns7, is named directly in two. The traits are mostly a program for noticing the other risks.

The nine traits of a product-minded engineer mapped onto the four risks: seven touch value risk, two usability, three feasibility, four viability Trait Value Usability Feasibility Viability 1 Proactive with ideas 2 Business and user data 3 Curiosity about “why” 4 Talks to non-engineers 5 Offers trade-offs upfront 6 Pragmatic on edge cases 7 Quick validation cycles 8 Owns results end to end 9 Instincts from repeated cycles Marks per risk (derived) 7 2 3 4 the trait text names the risk the writer's inference from the trait text
Figure 6. The nine traits of Orosz's product-minded engineer16 against the four risks7. The marks and the per-risk counts are the writer's reading of the trait text synthesis; Orosz makes no such mapping.

8.2 Behaviors as decision points

Two traits are concrete enough to be decisions. On trade-offs, the engineer looks both ways, for engineering trade-offs and for the product impact of cutting effort, and will "often go back to the product manager, suggesting a completely different feature to be built, given the product impact would be similar, but the engineering effort vastly smaller"16. That is a claim on rung 3 of Figure 1, the solution concept, made from the feasibility side. On edge cases the example is "if one in a thousand users might be hit by an error": weigh the effort to fix it against what happens if nothing is done, and ask "Can customer support help the person in this case, during validation? Can the user just retry and succeed the next time? Can the product be slightly modified, so this edge case won't occur?"16. Two of those three questions point at options that need no engineering work at all. The decision is a cut, which is rung 4 of Figure 1; Figure 4 shows it with builders in Scrum, PostHog and Linear.

The eighth trait fills the emptiest row of Figure 4. These engineers "consider their work done only after getting results on user behavior and business metrics", staying in touch with PMs, data scientists and support after rollout; "it can take weeks to get enough reliable data to draw conclusions"16. In the two models whose sources are silent on rung 8, Linear and Shape Up, this is how an engineer claims it: nobody is described as holding it, and the trait is a way to hold it. synthesis PostHog turns the traits into a job description with rights attached. Its "good product engineers" "ship quickly so they have a fast feedback loop", "understand the company strategy, and prioritize based on this and what they believe users want", are "good at descoping" and "make sure the things they've built are being used"15.

8.3 What an engineer can claim under each model

Table 4. Decisions the sources give engineers or builders under each model, and what the source asks in return. Rows are source statements regrouped by the writer synthesis; they describe product teams, and no source says they hold for an internal platform or runtime team.
ModelWhat the engineer holds or can claimWhat the source asks in return
CEO of the productThe "how"; the "what" only by persuading the PM10Build what the PM defines; accept that the PM will "manage the product team"
ScrumThe plan, the sizing and the method, "at the sole discretion of the Developers"; order only by convincing the product owner9A usable Increment each Sprint that meets the Definition of Done
TrioFeasibility, with the lead engineer "accountable for the product's delivery"7; a seat in solution discovery11; deference on technical calls2Answer for feasibility; be the party prototypes are tested with for feasibility2
PostHogWhat to build next, spec'ing solutions, sprint planning, and in stages 1 and 2 release timing and price1314; "the final call"12Track whether it is used15; accept the monthly review and challenge12; teams of six or fewer12
LinearLeading a project in rotation; scope inside a roadmap theme; readiness6Weekly update as lead; a 1-2 page spec; "do what nobody wants" in your turn6
Shape UpTasks and approach inside the pitch; a seat at the table for a senior programmer; shaping if technically literate3Ship inside the cycle: "by default the project doesn't get an extension"3

8.4 What it earns, and where the evidence stops

Orosz's claim about the payoff is an observation: such engineers "often become key contributors, the goto people for product managers and frequently advance to being team leads"16. The career payoff is asserted, not measured. PostHog's version is a grant of rights ("genuine autonomy and responsibility for product decisions") with no measured benefit12. Two limits apply to an inference-runtime engineer. Orosz places these engineers on "user-facing features" beside a PM16; the page says nothing about infrastructure or platform teams, and neither do the Linear and Shape Up pages. The only fetched statement about infrastructure-heavy work is PostHog's exception in Section 4.3, which asks for strong infrastructure skills in the lead and a PM early for direction, positioning and user research14. Whether Table 4 holds for a team whose users are developers embedding a runtime is not answered by any source gathered here; page 6 returns to the gap.

9. AI-era shifts in who does what

Page 1 (Section 8) covers the argument about what AI does to the cost of a prototype: that prototypes get cheaper and clear specifications gain value. This section covers only what the two statements imply for roles, and how little of it is measured. Both are practitioner assertions from the last two years and are likely to change within one.

9.1 The PM moves toward building, and the engineer is invited toward product

Cagan's 2026 FAQ says "the product manager is building and testing prototypes in order to learn whether the particular solution would generate the necessary outcomes", and that in discovery "AI plays more of a prototyping and decision support role"2. Ng's January 2025 letter says "increasingly, I also expect strong product managers to be able to build prototypes for themselves", and that "Software Engineers, being technical, are understanding and embracing AI much faster than Product Managers"17. His own sentence about the other direction is the one that bears on an engineer: "Perhaps engineers will step in to do some of this work, but if it remains the purview of specialized Product Managers, then the demand for these roles will grow"17. PostHog's PM team lists for Q2 2026 that every PM should track "human vs agent actions" and define "success for each", and every product team should state a "reason for being" in an "agent-first world"13. Read together, the statements move the PM's job toward the prototype, which was the engineer's tool, and leave open whether engineers take on specifying. synthesis None of the three offers a measurement of PM time, prototype cost or specification quality, and Ng's case rests on an economic analogy: cars and gasoline as complements17.

9.2 The one number: an Engineer:PM ratio with no source

Table 5. Every ratio between engineers and PMs found in the gathered material, with its basis. All Ng figures are unsourced assertions: the first row is from the fetched January 2025 letter; the two later rows are known only from search snippets that were not fetched.
ClaimValueBasisStatus
Ng: "many companies" (January 2025)176:1, range 4:1 to 10:1Assertion in a newsletter letterUnverified: no source or method given
Ng, later statement (undated here)down to 1:1Search snippet, not fetchedUnverified: search snippet only
One AI Fund team, as attributed to Ng1 PM per 0.5 engineersSearch snippet, not fetchedUnverified: search snippet only
PostHog, upper bound under its team capat most 12 to 24 team members per PMCap of six per team12 times two to four teams per PM in the handbook's list13Derived; a cap, not a headcount
Linear (October 2023)one PM in the companyEngineer's statement6Stated; headcount not given, so no ratio
Scrum Guide (2020)one product owner per Scrum Team of 10 or fewerTeams on one product "share the same ... Product Owner"9No fixed ratio

The derived PostHog bound sits above Ng's range and the sources do not reconcile the two: Ng gives a typical company ratio, the bound is a ceiling for one company's cap. Linear's figure shows that a very high engineer-to-PM ratio is possible for a small senior team, in that team's own account6. The honest summary of the field is that there is no sourced figure: one assertion in a fetched letter, two later figures known only from unfetched search snippets, and three structural descriptions.

10. The spec as a decision carrier

10.1 Five artifacts, compared

An artifact has to carry the upstream decisions (rungs 1 to 4 of Figure 1) in a form the people downstream can build from, at a level of detail that does not decide for them. Too little and "everyone – the product manager, the engineers, and the testers – may all assume different things"; too much and the product team "doesn't even read it", or it "becomes obsolete before it is even done"1. Shape Up adds an estimation argument: "the more specific the work is, the harder it can be to estimate"3. The five artifacts differ in where they start, who writes them, what a draft costs, and which risk they are built to reduce (Table 6).

Table 6. Five spec artifacts as their sources describe them. "Cost" is only what the source states; "not stated" means the fetched text gives no time or effort.
Artifact (source)Starts fromForm and lengthAuthorCost statedRisk it reduces
PRD (Cagan, 2005)1Homework on customers, competitors and the team's technology; a clear purposeTen steps; four areas: purpose, features, release criteria, schedule; a living documentThe product manager; designer on user profiles; engineers on feasibility tests"Depends greatly on the size and complexity of your product"Value, usability, feasibility, through step 5 prototype tests
Prototype as spec (Cagan, 2026)2A problem and an outcome, then a discovered solutionA prototype, plus a PRD of use cases and non-functional requirementsThe PM builds and tests prototypesNot statedAll four, each tested with a different party
Pitch (Shape Up)3A raw idea shaped within an appetiteFive ingredients; fat-marker sketches and breadboards, no wireframes or high-fidelity mocksShapers; Chapter 6 does not name the writerShaping runs on its own track beside the build cycle; no time givenRabbit holes (feasibility), scope (no-gos), problem fit
PR/FAQ (Bryar and Carr)4The finished customer announcementPress release plus external and internal FAQTypically a PM, solo in early draftsFirst draft "a few hours"; full process monthsValue, viability and feasibility, on paper
Project spec (Linear)56Collected feedback and open questions1-2 pages: "why", "what" and "how"The named project owner, a rotating engineerNot statedBuilding the wrong thing; lack of shared awareness

10.2 The 2005 PRD: the target in writing, the solution left to engineers

Cagan's paper calls the PRD "a more important, higher leverage piece of work" than any other a company produces, one that "drives the efforts of the entire product team and the company's sales, marketing and customer support efforts" (p. 1)1. Its ten steps run from homework (step 1) to managing the product (step 10), and its four areas are purpose, features, release criteria and schedule (pp. 9-11)1. Three features matter for the role question. First, the what/how line is explicit: a requirement should state "the need, rather than the solution", described "at the level of the interaction design and use cases" while "leaving as much flexibility to the engineering team as possible" (p. 10)1. The paper's own example is a performance one: ten customers describe ten different requests, but "they are all simply trying to get around the fact that an operation takes too long, so the root problem to be solved may be one of performance" (p. 15)1. A PM who specifies the solution "can frustrate and undermine the product development team", since "the solution is their contribution to the product" (p. 15)1.

Second, priority is a decision written into the artifact. Requirements are classed as must-have, high-want or nice-to-have, and a must-have "literally means that the product should not ship if even one of the 'must-have' features is not ready"; each class is then rank-ordered 1 to n, so that cuts under schedule pressure are decided in advance (p. 11)1. Third, release criteria give the engineer the readiness test that Figure 1 shows no other artifact carries: performance, scalability, reliability, usability, supportability and localizability (p. 10)1. Step 9 tests the document from the reader's side: "can an engineer get enough understanding of the target in order to get the product there?" and can QA design a test plan from it (p. 12)1. The paper's own warning is about the PM's confidence in the document: "one of the biggest and most common mistakes product teams make is to have far more confidence in their product specifications than they should" (p. 7)1.

10.3 Prototype as spec: the same author, twenty-one years later

The 2026 FAQ moves the primary carrier. Once a solution has been found in discovery, "the primary way we communicate what needs to be built by the engineers is via the prototype we created in product discovery. This is referred to as 'prototype as spec.'" What a prototype cannot show goes into a PRD: "an enumeration of specific use cases, and non-functional requirements (e.g. what are the scale expectations?)". The warning is about order, not format: "What is most important is that the PRD not be used instead of product discovery, in which case you are back to the project model"2. The same author's 2005 paper already had a prototype step, noted that prototyping tools let "engineers or designers rapidly create very functional prototypes" (p. 8), and listed release criteria that are the non-functional list1. So the change is one of primacy, not of kind: in 2005 the prototype tests the concept before the PRD is final; in 2026 it is the spec and the PRD is the remainder. For an engineer whose product lives on latency, memory and throughput, that remainder is most of the contract, and the FAQ's own example, scale expectations, is a number the prototype will not state.

One author's position on the PM and the spec in 2005, 2017 and 2026: PM paints the target and writes the PRD; PM responsible for value and viability with no artifact position; PM as builder with the prototype as the spec 2005 PRD white paper 2017 Four-risks essay 2026 Build to Learn FAQ PM's job Spec Prototype paints the target, writes the PRD value and viability risks builder, not manager; builds prototypes the PRD, ten steps is the spec no position on artifacts prototype is the spec; PRD supplements step 5, before the PRD is final not mentioned primary carrier of the solution
Figure 7. One author's position over three dates: the 2005 PRD paper1, the 2017 four-risks essay7 and the 2026 FAQ2. Dashed cells: the essay does not address artifacts. The 2005 paper was based in part on Horowitz's notes1, so the starting point is the CEO model of Section 2, not an independent view.

10.4 The pitch: a problem story, a time budget and a deliberately rough solution

Shape Up's pitch has five ingredients: problem, appetite, solution, rabbit holes and no-gos3. Each does a job. The problem is "a single specific story that shows why the status quo doesn't work", so the table has a baseline against which to test any solution; without a problem "there's no test of fitness to judge whether one solution is better than the other". The appetite "prevents unproductive conversations": "Anybody can suggest expensive and complicated solutions. It takes work and design insight to get to a simple idea that fits in a small time box"3. A problem with no solution "is unshaped work", and handing it to a team means "pushing research and exploration down to the wrong level"3. Rabbit holes and no-gos are the feasibility and scope notes: the first patches a risk the shaper found, the second marks what the concept will not do, such as the payment-form example that excludes WYSIWYG editing3.

The fidelity rule is the one that separates it from a PRD: "We need more concreteness, but we don't want to over-specify the design with wireframes or high-fidelity mocks. They'll box in the designers who do the work later"3. The pitch is read before anyone meets: posted so stakeholders "can read it on their own time", with comments "Not to say yes or no — that happens at the betting table — but to poke holes or contribute missing information"3. It carries rungs 1, 3 and 4 and by design leaves rungs 5 and 6 to the table and the build team (Figure 1).

10.5 The PR/FAQ: write the announcement, then list what must be true

The PR/FAQ has two halves. The press release has a heading, a subheading naming the customer and the benefit, a summary with a launch date, a problem paragraph "from the customer's point of view", solution paragraphs that must also say how the product differs from what customers use today, and two quotes. The FAQ has an external part, the questions customers and press would ask, and an internal part for "senior leaders and stakeholders", which "should cover all the challenging problems that need to be solved to build the product, whether technical, financial, legal, or operational"4. Page 1 (Section 6.3) covers why the authors call it a funnel and what they say it costs; this section covers its mechanics.

The mechanism is the order. A press release is normally the last step before launch, and "writing a press release is a forcing function to ensure that the creator of the new product idea is focused on the customer", not on "competitors, current capabilities, P&L, business partners"4. The page's worked example runs the test both ways. A release for a mid-size SUV with nothing better, faster or cheaper than existing ones "should be discarded before even writing an FAQ", because it "started with what your company can do". A release for a car with an 850-mile range, a 15-minute recharge and a $19,000 price is the right kind of release, although it "exceeds the technology, engineering, and cost of manufacturing abilities of not only your own company but every other car maker on the planet"4. The cars are the authors' hypotheticals. The rule drawn from them is stated outright: "If your PR doesn't require you and your team to solve at least one complex problem with a new and innovative approach, you should return to the drawing board"4. For an engineer this inverts the usual filter. Feasibility is not a veto on the release; it becomes the internal FAQ's list of problems to be solved. Among the five artifacts' sources, only the PR/FAQ page asks for viability numbers: "What are the per-unit economics of the product? What is our expected Gross Profit and Contribution profit per unit?"4. synthesis

1 First draft 2 Peers 3 First review 4 Executives 5 Decision 1 First draft 2 Peers 3 First review 4 Executives 5 Decision Solo author, typically a product manager. First draft: “a few hours, not a few days”. Early drafts have many holes the author must research. The author asks a manager and a few cross-functional peers. Fidelity and detail increase with each draft. A cross-functional group of about 10 contributors plus a decision-maker. 15-20 minutes of silent reading and note-taking. The remaining 40 minutes: questions, debate, discussion. The decision-maker shares it with the appropriate executives. If they find it promising: more drafts and more review meetings. Go: resources, a rough timeline and next steps are spelled out in the document. No-go reasons listed: copycat, TAM too small, payoff risky or low, an unsolved problem, or a backlog that ranks it later. Full process: months for the most successful Amazon products; AWS: more than a year.
Figure 8. The PR/FAQ process as the authors describe it, with their stated time costs: draft in hours, a first review meeting of about an hour (15-20 minutes of reading plus the remaining 40), and months to more than a year for the largest bets4. The five no-go reasons are the page's own list.
  1. Draft alone. One author, usually a PM, writes the press release and FAQ in hours; gaps are expected.
  2. Take it to peers. A manager and a few cross-functional peers fill holes; each draft is more detailed.
  3. First review. About ten contributors and a decision-maker read in silence, then spend the rest of the hour on questions.
  4. Executive reviews. A promising document goes to executives for further drafts and meetings.
  5. Go or no-go. A go carries resources and a rough timeline in the text; a no-go has five listed causes.

The review is built to find holes, not to win approval: "we were focused more on truth-seeking vs. selling and improving vs. deciding". The decision is a go or no-go once the document is complete enough, and a go carries "resources (people, money, fixed cost investments), a rough timeline, and other operational matters" in the text, with an executive staying in touch "with monthly or weekly reviews"4. Two limits. The page is by two former Amazon executives and sells a PR/FAQ course on the same site, so it is a practitioner account4. And it concedes the objection: "Outside of Amazon, most people have a profoundly negative reaction to spending weeks and months writing PR/FAQs before they start building a product"4; its answer is the distinction between speed and velocity, an argument and not a measurement.

10.6 The one-to-two-page project spec: written by the person who will deliver

Linear's method asks for brevity: "Short specs are more likely to be read. The purpose of a spec is to briefly communicate the 'why', 'what' and 'how' of the project to the rest of the team. Ideally these short documents force teams to scope out work so priorities are clear and teams avoid building the wrong thing"5. Its engineers describe the sequence: collect everything heard about the area in one place, "synthesize the common themes and open questions", run customer interviews, "once we have a rough direction, we document it, ask for a review, and iterate", and "aim to write concise specs (1-2 pages) outlining our key product and technical decisions"6. The stated purpose is awareness: "Giving people situational awareness is the main goal of project specs"6.

The difference from the other four is the author. The PRD and the PR/FAQ are written by a PM for others to build from; the pitch is written by shapers for a table and a team they are not on; the Linear spec is written by "a named owner who's responsible for writing the project brief and delivery"5. When the author is also the deliverer, the document does not need to transfer a decision across a hand-off. It aligns people who are already on the same side of it, which is why it can include the "how" that a PRD is told to leave out. synthesis The cost is the one the Linear account itself states: this works with "a fairly senior team"6.

10.7 Matching an artifact to a decision

Table 7. Which artifact the sources point to for each kind of decision, and what it leaves out. Rows are the writer's synthesis of Table 6 and Sections 10.2 to 10.6 synthesis.
Decision to carryArtifactWhyWhat it leaves out
Is this problem worth a team, and for whom?PR/FAQ, or a pitch's problem storyHours to draft; forces a named customer and a differentiation claim43Any customer reaction; the writer's belief against reviewers' skepticism
What shape of solution within a time budget?PitchAppetite, rabbit holes and no-gos bound the concept3Detail by design; who is on the team
Will the solution work for users and the business?Prototype, tested per riskUsers for value and usability, engineers for feasibility, stakeholders for viability2Scale, performance and other non-functional requirements
Scale, performance, reliability, release barPRD (release criteria)The only artifact here that states when to ship12The solution's feel; risks the prototype exposes
Who does what, how, inside a small senior team1-2 page project specWritten by the deliverer; carries "why", "what" and "how"56Anything about the decision to start

Where experts disagree

Two disagreements run through the sources: who holds the call on what gets built, and which document should carry the decision. Both are fought between organizations describing their own practice, so the comparison below is of mechanisms, conditions and costs, not of results.

Who decides what gets built?

The PM or product owner holds the call

Horowitz's memo10; the Scrum Guide9; Cagan's 2005 paper, which shares the memo's lineage1

Someone must answer for outcomes and must be able to say no, because engineers left alone drift to "whatever they want (i.e. solve the hardest problem)". Scrum gives one person the ordering and asks the organization to "respect their decisions".

  • One accountable voice: "one person, not a committee"9.
  • A rule against the engineering failure both older sources name101.
  • The memo is dated, and its host says it is "probably not relevant for today's product managers"10.
  • Eriksson reports the "mini-CEO" outcome8, and PostHog names the PM as "bottleneck and gatekeeper"12, both anecdotal.

Nobody holds it; defer by domain, PM accountable without authority

Eriksson8; Cagan 2017 and 202672; Torres11

The PM is responsible for value and viability, "not the boss of anyone", and the team "defers to the person that is best suited for the particular decision". Leaders pick the problem; the team solves it.

  • Fits the stated decision rate: a single decider saturates (Section 3.3, derived).
  • Keeps accountability while dropping command.
  • Names no tie-break and no list of which domain belongs to whom.
  • "Earned authority" gives a new PM no rule on the first day.

Engineers decide; the PM is a compass

PostHog essay and handbook1214; Linear6; Shape Up's build-team autonomy3

"Ultimately it's the engineers who make the final call"; the PM supplies context, owns a few domains and runs the review. At Linear engineers lead projects in rotation with one PM.

  • Removes the gate; the people with the most technical information decide.
  • A written tie-break for crossing calls: the "informed captain"14.
  • Conditioned on trust, teams of six or fewer, and demand others have shown1215.
  • PostHog hired PMs, and lists the "accountability gap" as open13; Linear's account is three years old and silent on PM rights.
Assessment

The positions answer different conditions, and none has outcome evidence. The arithmetic in Section 3.3 rules out the strong form of the first: at the decision rates the 2026 FAQ states, one decider cannot take most decisions, so what remains contested is who holds rung 1 (what to build next) and rung 8 (did it work). Where strategy is strong and engineers drift to interesting problems, the first position's failure mode is the live one and a PM or product owner with a clear remit is the remedy. In small, senior, high-trust teams building things others have already shown to be wanted, the third position is credible, but PostHog's own reversal says the number of products matters. The second is the sensible default in between, if the team writes down its domains, which is what Figure 4 attempts. Evidence that would settle it: outcomes for matched teams under different models, or a company publishing before-and-after results when it changes model. None exists in the sources.

Which artifact should carry the decision?

A PRD, written to be complete

Cagan, 20051

"It's hard to come up with a more important, higher leverage piece of work for a company"; the document states purpose, features, release criteria and schedule, rank-ordered.

  • The only artifact here that states release criteria and priority order.
  • Step 9 tests it from the reader's side: can an engineer and QA work from it?
  • Its author later calls the PRD the remainder2, and the 2005 paper itself names over-confidence in specifications a common mistake.

The prototype, with a PRD for what it cannot show

Cagan, 20262; partly Ng17

Discovery produces a tested prototype; that is the spec, and the PRD lists use cases and non-functional requirements.

  • Tests each risk with the party who can answer it2.
  • Cheaper prototypes make the route plausible, per both authors.
  • Scale expectations cannot be read off a prototype.
  • Ng expects clear written specs to gain value17; the 2026 FAQ names no AI-specific risk of prototyping2.

Short written narratives, each for one decision

Shape Up's pitch3; the PR/FAQ4; Linear's 1-2 page spec56

Keep the artifact rough (pitch), customer-first (PR/FAQ) or brief enough to be read (spec).

  • Cheap to write; matched to one decision each (Table 7).
  • The pitch refuses to over-specify; the spec is written by the deliverer.
  • The PR/FAQ is months of work for large bets, per its authors4.
  • The Linear spec leans on "a fairly senior team"6.
Assessment

These are less rivals than answers to different decisions (Table 7). The PRD and the prototype both carry a solution to builders who are not its authors; the pitch and PR/FAQ carry a decision to a forum; the project spec aligns people already on one side. The real dispute is whether the carrier is a document or a running thing. For a product whose contract is non-functional, such as latency or memory on a target device, the 2026 position itself sends that contract back to a written list, so a runtime team needs the written part regardless. What would settle the rest: rework rates or defects traced to spec gaps across artifact types. No source measures them, so every claim of superiority here is an author's belief.

What is changing

Three things on this page are likely to move within a year. PostHog's handbook is a living document: its product-team page already carries two PM counts and an open "accountability gap" theme13, and its stage rule and infrastructure exception are policy as of 2026-10-2614. The AI-era role claims from Cagan and Ng are weeks to months old and unmeasured217. Linear's single-PM statement predates this page by three years6. The sources also span three decades, which is why the chronology matters when weighing agreement.

  1. CEO-of-the-product memo written

    Derived from the host's "15 years ago" in 201210.

  2. Cagan's PRD paper

    Based in part on Horowitz's notes1.

  3. Memo posted with a relevance warning

    The host's own caveat10.

  4. Eriksson (March 15) and Cagan's four risks (December 4)

    The "not the CEO" and trio statements87.

  5. Orosz on product-minded engineers

    Nine traits, no data16.

  6. Scrum Guide 2020

    Product owner accountabilities9.

  7. Linear on rotating project leads

    One PM at the time6.

  8. PostHog essay

    Engineers run the teams; PMs as compass12.

  9. Ng on AI product management

    Ratios and prototype-building PMs17.

  10. Cagan's Build To Learn FAQ (April 27) and PostHog's Q2 handbook themes

    Prototype as spec; agent-first success definitions213.

References

  1. blogCagan, M. How To Write a Good PRD. Silicon Valley Product Group white paper, 26 pp., © 2005 (PDF re-hosted 2024-07). https://www.svpg.com/wp-content/uploads/2024/07/How-To-Write-a-Good-PRD.pdf. Accessed 2026-10-26. Used for: ten steps and four PRD areas (pp. 9-11), prioritization (p. 11), step 9 completeness test (p. 12), prototype step and over-confidence (pp. 7-8), what versus how (p. 15), engineering-driven requirements (p. 17), Barksdale and Netscape (p. 16), footnote 2 on Horowitz's notes (p. 1).
  2. blogCagan, M. Build To Learn FAQ. Silicon Valley Product Group, 2026-04-27. https://www.svpg.com/build-to-learn-faq/. Accessed 2026-10-26. Used for: PM not "the decider", surgical-team deference, builder not manager, strategy picks the problem, the implicit agreement, prototype as spec, PRD role, who each risk is tested with, AI in discovery, "dozens or even hundreds of decisions every day".
  3. officialSinger, R. Shape Up: Stop Running in Circles and Ship Work that Matters. 37signals / Basecamp, web edition (no edition year on the page). https://basecamp.com/shapeup. Accessed 2026-10-26. Used for: Ch. 1 audience list; Ch. 2 level of abstraction, who shapes and the two tracks; Ch. 6 pitch ingredients and fidelity; Ch. 8 betting-table membership, team composition and circuit breaker; Ch. 10 assigning projects not tasks and "done means deployed" (Ch. 10 fetched beyond the source map).
  4. textbookBryar, C. and Carr, B. The Amazon Working Backwards PR/FAQ Process. workingbackwards.com, authors' companion site for the book Working Backwards (undated page; the book text was not consulted). https://www.workingbackwards.com/concepts/working-backwards-pr-faq-process/. Accessed 2026-10-26. Used for: press-release and FAQ components, review meeting format, process stages and time costs, decision outcomes, the car examples, the standard internal FAQ questions (Appendix 2). The page sells a PR/FAQ course.
  5. officialLinear. Linear Method: Principles & Practices. Linear, undated. https://linear.app/method/introduction. Accessed 2026-10-26. Used for: project owner rule, "Write project specs", cross-functional teams; the page never mentions product managers.
  6. blogYu, N., Roman, S. and Largman, R. How we run projects at Linear. Linear, 2023-10-05. https://linear.app/now/how-we-run-projects-at-linear. Accessed 2026-10-26. Used for: engineers leading projects with one PM, rotating leads, what counts as a project, roadmap themes, 1-2 page specs, readiness decided by engineers, the size-and-seniority condition.
  7. blogCagan, M. The Four Big Risks. Silicon Valley Product Group, 2017-12-04. https://www.svpg.com/four-big-risks/. Accessed 2026-10-26. Used for: PM, designer and lead engineer as responsible and accountable for named risks.
  8. blogEriksson, M. Product managers — you are not the CEO of anything. Mind the Product, 2017-03-15. https://www.mindtheproduct.com/product-managers-not-ceo-anything/. Accessed 2026-10-26. Used for: no direct authority, responsibility conceded, Ken Norton quote, team sport, the mini-CEO failure.
  9. officialSchwaber, K. and Sutherland, J. The 2020 Scrum Guide. scrumguides.org, HTML port of the November 2020 version. https://scrumguides.org/scrum-guide.html. Accessed 2026-10-26. Used for: Scrum Team, Developers and Product Owner accountabilities, Sprint Planning topic three, sizing, team of 10 or fewer, shared Product Owner (fetched beyond the source map).
  10. blogHorowitz, B. Good Product Manager/Bad Product Manager. Andreessen Horowitz, posted 2012-06-15 with the host's warning that it was written 15 years earlier. https://a16z.com/good-product-manager-bad-product-manager/. Accessed 2026-10-26. Used for: the CEO analogy, the what/how split, management of the product team, written positions, the engineering failure mode, the date warning.
  11. blogTorres, T. Opportunity Solution Trees. Product Talk, undated. https://www.producttalk.org/opportunity-solution-tree/. Accessed 2026-10-26. Used for: definition of the product trio and the tree as a team activity.
  12. blogHawkins, J. and Neu-ner, L. Product management is broken. Engineers can fix it. PostHog, 2024-12-03. https://posthog.com/newsletter/product-management-is-broken. Accessed 2026-10-26. Used for: failure mode, engineers run product teams, final call, the session-replay story, the reversal and "8+ products", trust, team cap of six, the growth-review split of rights.
  13. officialPostHog. Product management at PostHog. PostHog Handbook, accessed as a living page with Q2 2026 themes. https://posthog.com/handbook/product/product-team. Accessed 2026-10-26. Used for: engineers own sprint planning and spec'ing, PM duties, 1 interview per week per PM, the PM-to-team listing (derived counts), the team with no PM, Q2 2026 themes including the accountability gap and agent-first goals.
  14. officialPostHog. What product managers do at PostHog. PostHog Handbook, living page. https://posthog.com/handbook/product/product-manager-role. Accessed 2026-10-26. Used for: own, share and support tiers (Table 1), team lead as closest to product owner, informed captain, stage table, when to add a PM, the infrastructure-heavy exception (added beyond the source map).
  15. officialPostHog. How to do product, as an engineer. PostHog Handbook, living page. https://posthog.com/handbook/engineering/product-engineering. Accessed 2026-10-26. Used for: good and bad product engineers, pre-PMF build-only-proven-demand rule, "just shipping", staggered rollouts, engineers booking their own interviews (added beyond the source map).
  16. blogOrosz, G. The Product-Minded Software Engineer. The Pragmatic Engineer, published 2019-09-05. https://blog.pragmaticengineer.com/the-product-minded-engineer/. Accessed 2026-10-26. Used for: nine traits, the trade-off and edge-case examples, end-to-end ownership, the career claim, the user-facing-features setting, secondhand quotes from an Atlassian PM and Shopify's head of engineering.
  17. blogNg, A. The Batch, Issue 284: letter on AI Product Management. DeepLearning.AI, 2025-01-15. https://www.deeplearning.ai/the-batch/issue-284. Accessed 2026-10-26. Used for: Engineer:PM ratio assertions (6:1; 4:1 to 10:1), PMs building prototypes, engineers adopting AI faster, engineers possibly taking on PM work.