The product manager job AI wants is the one from 1931

In 1931, a junior executive at Procter & Gamble wrote an 800-word memo asking for two more hires. He wanted someone accountable for each soap brand: someone who tracked sales, ran the promotions, handled the packaging, and talked directly to the people buying the soap. P&G gave him the hires, and the product manager job came out of that request. Ninety-five years later, AI is stripping the role back down to exactly that shape, mostly by deleting twenty years of ticket-writing that a different job was supposed to own in the first place.

A memo from 1931 already described the job

Neil McElroy was the junior executive, and the problem he was solving was internal competition: two P&G brands, Ivory and Camay, fighting over the same customers with no one accountable for either. His memo proposed a role he called brand man, responsible for a product’s full outcome: sales tracking, packaging, promotion, and direct field contact with the people buying it. P&G gave him his two hires, and modern product management started from that decision.

The job traveled well past soap. McElroy later became president of P&G, and every P&G president since has come up through brand management. He also advised two Stanford engineering students named Bill Hewlett and David Packard, and HP built its own product manager role in the 1940s on the same logic: one accountable owner sitting closer to the customer than the engineering managers building the thing.

That’s where the mini-CEO description comes from. People still argue about how literally to take it, but the accountability behind it wasn’t rhetorical. Read what McElroy actually asked for, and the job has almost nothing to do with backlogs, sprints, or story points. What he asked for was market knowledge and enough standing to make a call and answer for how it turned out.

Scrum invented a narrower job and the title swallowed it

The engineers who wrote the Agile Manifesto in 2001 were solving a real problem: software shipped slowly, requirements arrived from on high, and nobody got to test whether any of it worked before it went out. Scrum, built out of that manifesto, added a role called product owner: responsible for the backlog, the sprint, and turning stakeholder requests into stories a team could build. It was a narrower job than the one McElroy described, a role built to solve a coordination problem inside a single team, with a much smaller mandate than owning a product’s fate in a market.

What happens when the two roles get blurred together is fairly predictable, and I’ve watched it play out in enough product orgs to recognize the shape early. Product managers write requirements and hand them down to product owners who have never spoken to a customer. Product owners write stories and hand them to teams that never get to check whether any of it worked. Every handoff looks like progress on a board somewhere. Nobody validates anything, and the job turns into a relay of paperwork with a market outcome attached to the end of it by accident.

At scale, the SAFe framework made the split official, ranking product manager above product owner as an external-facing role that hands requirements down to an internal-facing one. That’s a role built on market judgment turned into a reporting hierarchy. Plenty of people carrying the title product manager over the last twenty years spent most of a working week grooming backlogs, writing tickets, chasing dependencies, and running ceremonies that a different job description was supposed to own.

Same title, three different jobs

Line up the job across three periods and the pattern holds together on its own.

AI is deleting the work Agile added to the job

What’s happening to the role in 2026 isn’t complicated. Backlog grooming, ticket writing, status updates, and dependency tracking are migrating into AI agents running inside Jira, Linear, and the MCP integrations connecting them together. One widely read breakdown of this year’s hiring data put it plainly: teams are getting leaner, product managers are owning more surface area, and the project coordinators and scrum masters who used to absorb this work are getting folded into product roles or cut outright, because their lowest-judgment work now belongs to software. In the strict Scrum sense, product owner is the single most exposed title on a product team right now.

The busywork is what AI is removing: grooming, ticket-writing, status updates, dependency tracking. Deciding what’s worth building, and getting a room of people and agents to commit to it, is the work nobody has managed to automate, because it requires actually understanding the market a decision gets scoped for.

The product manager role isn’t the only one moving. The line between product managers and product designers has blurred faster than most people expected, since designers are now using AI to build functional prototypes and product managers are using design tools to mock up flows without pulling in a designer at all. That blur is part of why founders at smaller companies are experimenting with a combined product engineer role instead of keeping the two separate. Product and data analysts, who own instrumentation and the quantitative side of a decision, may be the most underrated seat left standing, since the judgment this whole shift protects still runs on numbers somebody has to instrument.

Hiring managers are already paying for the difference

AI has made specialization more valuable, which is the opposite of what a lot of people expected from a technology that supposedly flattens expertise. Tuning a large model behaves nothing like shipping a consumer feed. A PM with real depth in either one keeps outperforming a generalist trying to cover both, and the job postings have shifted to match: fewer openings for a generic product manager, more for an AI product manager, an API product manager, a consumer product manager.

Compensation is following the same line. The published 2026 US figures put AI-focused product managers at roughly double what generalist PMs earn, a gap wide enough that it’s hard to explain as noise. Hiring managers are paying for people who know a domain well enough to make the call and defend it, which is the same thing P&G was paying for in 1931.

Agile is not the thing that’s ending

None of this makes Agile a mistake: plenty of teams still get real value from a shared rhythm for turning a decision into shipped work, and their scrum ceremonies aren’t going anywhere. The narrower thing ending is two decades of people carrying a product manager title while spending most of the week doing a project manager’s job, a distinction nothing forced until AI started doing the ticket-writing itself.

A decision still has to reach whoever builds next

Deciding what’s worth building only helps if the decision reaches whoever builds it next, human or agent. That’s the part of this shift I keep seeing teams underestimate. A product manager doing the 1931 job well still has to pull signals out of Gong calls, Zendesk tickets, Salesforce notes, and Slack threads before anyone can act on it. Without that step the same problem just moves downstream, to an agent that can build a feature in a day with nothing in the stack telling it which feature was worth building.

Why we built Bagel around that gap

This is the gap we built Bagel to close. Bagel is a product brain: it reads the customer signal already sitting in Gong, Salesforce, Zendesk, Slack, and Jira, resolves it to one canonical customer record, and turns it into a scoped, revenue-weighted decision before a person or an agent has to sort through the raw noise by hand. An agent re-reading an unscoped signal lands around 21 percent accuracy, by Anthropic’s own research. Scoped first, that number moves to 95 percent plus.

One of our own people tested the alternative directly: our Head of Product Growth spent 50 hours of his own time rebuilding Bagel from scratch with Claude Code, and got remarkably far. That’s the honest answer on build versus buy. The concept isn’t hard to prototype over a weekend. Getting entity resolution, drift, SOC 2 and trust right at the scale a real customer base runs at is the part that takes longer.

Ohad Biron is the Co-Founder and CEO of Bagel AI – the first AI-native Product Intelligence platform. He’s spent the last decade building in the space between revenue, product, and customers. His mission: eliminate guesswork from product decisions and help teams ship what actually matters.