PM taste - 99% should-we, 1% can-we

PM as 99% discrimination: the cumulative weight of refused features, NFRs as primary work, and the featherless-hat invisibility of the craft.

This is one of twelve themes in the wiki. It covers what PM work actually is: not roadmaps, not meetings, not feature-shipping - but the discipline of discrimination at scale. You are likely here because you encountered the 99/1 ratio or the featherless-hat metaphor and want the full argument. This page sits between spec-first-taste and AI PM skillset: the foundational craft dispositions that predate the AI turn, and that the AI turn only sharpened.

One setup note: every belief here is held with conviction AND held provisionally. The meta-belief from 2021 - "I have a very strong opinion about having no strong opinions" - is the license. It explains why strong positions sit next to acknowledged scope limits.

The meta-belief

On 2021-12-12, one sentence: "I have a very strong opinion about having no strong opinions."

This is the frame. The PM who holds no opinions says yes to 99 of 99 feature requests. The PM with only rigid opinions calls stubbornness taste. The productive space: conviction held hard, discarded faster when evidence contradicts it. A PM without strong opinions is a pushover. One who cannot update them is a bottleneck. The ten beliefs on this page sit inside that tension. Each is held with conviction. Each has an acknowledged scope limit. Cross-listed in linkedin-as-instrument as the disposition that licenses public takes on contested topics.

What PM work actually is

Product management is not a role you see when it is done right.

The 2022-06-03 post put a ratio on it: "Product management is the grind of answering 99 questions of 'should we do it?' to get to that 1 bonus question of 'can we do it?'. That 1 is disruption. The rest, food." Every day is 99 should-we judgments. The rare can-we only materializes when the 99 were answered well. Say yes to 90 of the 99 and the pipeline fills with features that defer the one question worth asking.

Ten weeks before that post, the output of those 99 decisions got a name: the featherless hat. From 2021-09-22: "Product management is a featherless hat that is omnipresent yet indistinguishable if done properly." A feathered hat announces itself. A featherless hat does the job without claiming the room. PM work touches story-level QA, PAT, business-requirement framing, stakeholder alignment. When it works, the product foregrounds engineers and designers. When it fails, every user notices. Two days after that post came the counterbalance: "I truly believe it's near impossible to hide your performance. If you're good at it, no amount of politics will hold you back." The artifact is invisible. The track record compounds.

Before the ratio and the hat, there was parenting. From 2021-07-12: "Building a product is like raising a child. You'll have to make tough decisions and piss people off at higher places to ensure least influence on your goal." Refusal as PM craft. Not creation. Not shipping. Three metaphors in a tight cluster - parenting (July 2021), featherless hat (September 2021), 99/1 ratio (June 2022) - each addressing the same thing at a different register. Parenting is the emotional posture. The hat is the output. The ratio is the cognitive work.

The data literacy layer

The PM who cannot scope their data's explanatory power is doing astrology with better charts.

The 2022-05-17 post - 78 reactions, highest-engagement craft post of that era - put it without commentary: "Just because you open an umbrella when it's raining does not mean it will rain whenever you open the umbrella. Not everything can be explained by your data. Identify what can be, and build accordingly." Data literacy is the discipline of knowing which questions data can answer - then building against that category. The PM who cannot draw that boundary will correlate their way into the wrong product. Some should-we judgments are answered with data. Others require taste. Knowing which is which is the craft.

The specification layer

Non-functional requirements are where most products fail. Performance, scalability, security: not afterthoughts delegated to QA. The specification layer where the PM does the work that no one credits when it holds and everyone notices when it breaks.

From a 2024-03-08 Collab Article: "Non-functional requirements are not just tick boxes that a PM needs to go through but actively work towards." NFR targets - latency thresholds, uptime requirements, security scope - are specified upfront, owned by the PM. Feature completeness is table stakes. When uptime holds, no one credits the PM. When it fails, every user notices. Featherless-hat work at the requirement layer. Invisible when right. Audible when wrong.

Design thinking lives at this same layer - as a speed tool. From 2024-05-24: "Don't confuse design thinking with making UI/UX first. It is the process of aligning stakeholders and gathering diverse feedback." The paper wireframe and the model-comparison notebook are the same move: make something touchable before committing resources. The build-measure-learn loop is only worth running if the first build is aimed correctly.

Taste as a personal, costly commitment

The anti-customization post from 2025-12-04 is the clearest first-person expression of PM taste as a held, costly stance.

"I am extremely opinionated about adding customization to products. To the point where I feel like I've lost a hard-fought battle with myself if a PRD ends up including customization settings."

The 2021-22 metaphors described PM work from the outside. This one describes it from the inside: a battle with the self. "Super customizable" sells in the meeting room. Taste refuses it anyway. Each customization setting is a design decision deferred to the user - a pre-paid mortgage on every future deployment. This is the 99/1 ratio made personal.

One scope limit: enterprise-rooted. The economic argument - six months of implementation, an implementation partner, hundreds of training documents - is shaped by enterprise deployment cycles. The conviction has not been tested against consumer products where the cost structure differs.

The compound

All of these beliefs run the same direction.

Refuse before you specify: the 2019 seed post, "Learn to say no. Learn when to say no. Learn who to say no to." Measure what the data can actually explain: the umbrella post. Treat NFRs as first-class work, not sprint-end QA. Align before building: the paper wireframe. Ship with conviction, not configurability: a product with no settings pane ships a PM's decided stance; 47 preference toggles ship a PM's inability to decide.

Taste is the cumulative weight of these moves, repeated until they become reflex.

Where to go from here

Three exits, depending on what you came for.

If you want the builder-stance application - how these discrimination instincts carry into working with AI tools - read spec-first-taste. The April 2026 trilogy (Spec > Sprint / Taste > Execution / Context > Prompt) is the AI-era version of the same beliefs. Same taste, different surface.

If you want the craft in applied domain - how PM taste maps onto the AI PM role specifically - read AI PM skillset.

If you want the meta-belief in its own context - the strong-opinion-about-no-strong-opinions productive paradox and why it licenses public posting - read linkedin-as-instrument. The 2021-12-12 anchor cross-lists there.