Cagan's new rule retires his own model next

A product manager who takes a roadmap of features with due dates and ships them is, in Marty Cagan’s words, “not a product manager at all.” They “happen to sit in the product org.” [reported] Cagan, the most widely read voice in product management and the founder of Silicon Valley Product Group, said it from the keynote stage at PendomoniumX, a product conference, this August. The title says product manager. The work is project management.

If you’ve spent a few years in product, you know why the line lands. Cagan has spent fifteen years arguing that most companies run feature factories, and that real product work looks nothing like one. This keynote hardened that argument into a definition. The bad version doesn’t get graded as weaker product management. By this definition it isn’t product management at all.

Two terms carry the whole keynote, so here they are in plain words. A feature factory is an org where someone hands the PM a prioritized list with due dates, and the PM’s job is to get it built and report progress. An empowered team is Cagan’s alternative: the PM is handed a problem to solve, gets measured on whether the business outcome moves, and works as a co-equal with a designer and an engineer. Cagan calls those three the product trio. Discovery, the ongoing work of figuring out what’s worth building, runs continuously alongside delivery. The trio is the thing Cagan is defending. It’s the good version.

Then he stopped. He pointed the definition backward, at the model empowered teams replaced, and never turned it forward, at the model replacing empowered teams. If the role is the operating model, the empowered-team PM is no more permanent than the feature-factory PM he just excommunicated. The same rule retires both.

Start with the rule itself, because it’s stronger than it sounds. Cagan isn’t saying good PMs adapt to their operating model. He’s saying the operating model constitutes the role, so completely that a feature-factory PM and an empowered-team PM are two different professions wearing one title. Their competencies are “nearly orthogonal,” in his words. [reported] Grant him that. It’s clean, it’s falsifiable, and I think it’s mostly right.

A rule like that has no off switch. It cuts in whatever direction you point it. Cagan aims it at the past, where it demotes the pipeline-manager to a project manager in disguise. Aim it at the future and it does the same work on his own model, because the next operating model is already here.

That model is the AI-native one, and the plain version is this. AI can now do the execution a team used to do: write the code, build a working prototype, produce the design mocks. When execution gets that cheap, the reason you needed three specialists in one room starts to go away. Coinbase, the crypto exchange, has been rolling out what it calls AI-native pods, including one-person product teams, where a single contributor runs a fleet of AI agents and shapes the UX and the final mockups alone. [reported] A two-person team there reportedly shipped in weeks what used to take months. [reported]

Hold that against the trio, and the problem is immediate. The empowered-team PM isn’t defined by doing discovery. Plenty of roles touch discovery. The definition is being a co-equal in a group of three. Co-equality is a relation. It needs at least two other people to hold it. When one contributor spans what used to take a PM, a designer, and an engineer, the trio doesn’t get harder. It loses the other two chairs. A trio of one has no one to be co-equal with.

The economics say the same thing without the drama. A trio is a specialist-heavy structure, and specialists carry a coordination cost. That cost was worth paying when writing code and producing designs was slow. Once AI makes two of those three crafts cheap, the coordination between them stops buying you anything. The structure Cagan uses to define his good PM is the structure the successor model erases.

The trio is one defining piece. Continuous discovery is the other: the patient problem-framing, the hypothesis-testing, the consensus-building across the team. And here Cagan has already published the finding that guts it. In his own SVPG writing this summer, he described how the skills that make a great discovery-model PM become “an active liability” in an AI-native one. Careful specification and consensus-building make you “too slow” when a working version can be built almost instantly. He even put a number on it: the share of a PM’s time spent aligning everyone before a build collapses from 40 to 60 percent down to under 15. [reported]

Read that next to the keynote, and the contradiction is his own. His case against the feature-factory PM is that their core work, running a delivery pipeline, is the part AI automates, so they’re doing the commodity instead of the craft. Fine. The empowered-team PM’s core work, trio facilitation and continuous discovery, sits on the automated side of that same line, by his own account. Cagan sharpened a blade against feature factories: your defining work is the part that automates. He never checked whether the other edge was pointed at his own model.

The honest objection is that Cagan did name AI-native. He isn’t blind to it. The keynote lists it as a third archetype with its own skills: ideation, rapid prototyping, constraint navigation. [reported] Isn’t that the forward-looking move I’m demanding?

Not quite, and the gap decides everything. Cagan files AI-native as a peer of the empowered team. A fourth quadrant. A sibling archetype sitting alongside the others, one a PM picks based on their org. The evidence describes something else: a successor that dissolves the two properties defining the empowered-team model, the trio and continuous discovery. You can’t be a co-equal sibling to the thing that absorbs you.

Which leaves the question Cagan’s own frame produces and then refuses to ask. When execution collapses the trio into one person, is that person still a PM?

It’s answerable, and the early evidence points somewhere uncomfortable. Watch who the one-person teams actually are. At Coinbase, the survivor manages agent fleets and shapes the mockups. If the trio collapses into the engineer, and the person left standing gets hired and paid as a builder rather than a PM, Cagan’s succession runs the opposite direction from his preference. The successor to the empowered-team PM might not be an AI-native PM. It might be someone who was never a PM at all, reclassified the same quiet way the feature-factory pipeline-manager just was.

So if you’re taking Cagan’s advice, take the true half and keep watching the rest. The true half is solid: get out of the feature factory. The rest is the catch. The empowered team is a way station, and continuous-discovery facilitation is a skill with an expiration date under the model that’s arriving. The thing worth compounding sits underneath all of it. Deciding what to build. Proving it worked. Knowing when to stop. That judgment outlives the trio, because it never needed two other specialists in the room to exist. Cagan is right that the operating model constitutes the role. He’s just one operating model behind on which one is constituting you.