For Marty Cagan, AI doesn't end the PM role: it ends two types of PM

Every founder has heard the two contradictory versions: "the product manager role is dying" and "product manager will be the most important role of the AI era." In a keynote for the Product Leader Summit 2026, published as text on November 6 on the Silicon Valley Product Group website, Marty Cagan gives the simplest explanation I've seen for this contradiction: both statements are true, because there are three different roles hiding under the same title, and AI is just accelerating a divorce that should have happened already.

Cagan's thesis is direct. What defines a PM's job isn't the company, the product, or the industry: it's the operating model. Companies that run on the project model (a roadmap of prioritized features, with the solution already prescribed) produce two roles: the feature team PM and the Scrum or SAFe product owner. Companies that run on the product model (a roadmap of problems to solve, with an open outcome) produce a third: the empowered team PM. These are different crafts, with different skills, and that's where the problem lies for anyone hiring or building this team today.

The Two Roles Automation Is Already Erasing

The feature team PM, according to Cagan, was never really a product manager: it was a project manager in disguise, whose job was to coordinate resources to deliver a solution the roadmap had already decided on beforehand. It's the PM who writes documentation one day, does QA the next, and tweaks simple front-end work the day after that. This is exactly the kind of work agentic AI is automating first, because companies are rushing to build what Cagan calls a "software factory": agents doing what used to be the job of QA, release managers, and this type of PM.

The agile squad product owner faces a similar risk, for an even more specific reason: their job is to manage the delivery backlog and coordinate sprints. If AI speeds up and automates much of the delivery process itself, the ritual around which the product owner organizes their day loses its purpose. Cagan is direct: "the dedicated product owner role is going to resolve itself," and he has been recommending that these people upskill as soon as possible.

For anyone founding or leading a tech company, this isn't an academic debate over naming. If your product team today spends most of its time on sprint ceremonies, backlog prioritization, and coordinating delivery of already-defined features, you're paying a PM's salary for work that the market's own dynamics are making redundant, regardless of what the job title says.

The Role That Survives, and Why It Gets Rarer

The empowered team PM is a different thing. Their core responsibility isn't delivering a prescribed solution, it's helping discover which solution solves the problem and delivers the promised outcome. That requires what Cagan calls product discovery: testing prototypes with users (value and usability), with stakeholders (business viability), and with engineers (technical feasibility), using qualitative and quantitative techniques, not meeting-room guesswork.

This is the part of the job AI still doesn't do well, because it isn't a defined workflow, it's judgment. Cagan notes that AI helps with this process (understanding market dynamics, competition, customer behavior), but as a tool that supports product sense, not a substitute for it. That's why this is the only one of the three profiles in high demand right now, according to him.

Why This Matters Even More in AI Products

This is where the point that matters most for anyone building a probabilistic product comes in, that is, a product whose behavior depends on a model and isn't deterministic. Cagan argues that the empowered team PM needs to be at the center of the eval process, the definition of what counts as a good or acceptable result, precisely because the viability risks (unit economics, legal liability, security, privacy, ethics) are bigger and less predictable than in conventional products.

Translating for anyone raising a round or deciding on headcount: if your product uses generative AI at its core, outsourcing the "is this good enough to ship" decision to engineering alone or to an agent alone means delegating exactly the kind of viability judgment Cagan says is the core of the role that survives. It's not a task for automation, it's a task for accountability.

The Signal Already Showing Up in Hiring

Cagan cites a recent observation from Shreyas Doshi, which sums up well what's already changing in hiring processes:

Product Managers who've historically viewed their role as "my main job is to execute" are finding it way harder to get hired at top companies than those who've viewed their role as "my main job is to make the product successful". For a couple of decades now, many companies (and leaders) were explicitly or implicitly giving PMs the unfortunate message that a PM's main job is execution. The truth is that execution, while extremely important, was never a PM's main job. Enabling product success was always the main job.

Shreyas Doshi

If you interview PM candidates, this is a better cutoff question than any framework case study: ask the person to describe the last time they killed their own feature because of discovery data. Anyone who can only describe deliveries is, by Cagan's definition, a product owner with a fancy title on LinkedIn.

The Counterargument You Can't Dismiss

Cagan himself mentions that "some important industry leaders proclaim that the product role is dead." It's worth taking this view seriously, because there's some truth to it: if AI already helps understand market and competition today, it's reasonable to bet that, at some point down the line, it will be able to run larger parts of user research on its own, further flattening the need for a human in the middle of discovery.

The weak point of this view, to me, is treating discovery as information gathering. It isn't. It's judgment about business viability, legal exposure, ethics, and strategic trade-offs, decisions that carry an accountability a company can't outsource to a system with no formal accountability. An AI agent doesn't sign a liability agreement with an investor, nor does it answer to a privacy lawsuit. As long as that remains true, the empowered team PM stays the name the company points to when a product decision goes wrong, and that weight is what keeps the role relevant even as the tooling around it changes every year.

The Question to Ask About Your Own Team

Before cutting or hiring a PM thinking you're "adapting to AI," look at what the person actually does day to day. If their job is mostly coordinating delivery of something the roadmap has already decided, you're paying for a role the industry itself is moving away from needing, with or without AI. If their job is deciding, with evidence, what's worth building and taking responsibility for that judgment, you're protecting exactly the part of the product that no agent replaces yet.