Product Management in the AI Era: What Should PMs Build and Own?

Imagine customers repeatedly abandoning a sign-up form. A product manager (PM) notices the problem, investigates the cause, and works with the team on a solution. Once the improved experience is released, the PM also needs to find out whether fewer customers are dropping off.
AI can now help with more of that work: sorting customer inquiries, drafting screen layouts, and even writing code for a prototype. Tasks that once required requests to several specialists can increasingly be tried by the PM directly.
How far should a PM’s role expand? Does that mean taking over the work of designers and engineers? Drawing on a November 12 Lenny’s Newsletter meetup in Seoul, where PMs working with product teams in the US and Japan shared their experiences, let’s explore the role and responsibilities of a PM in the AI era.
A PM’s work starts with the problem

When you think of a PM, you might picture someone writing a product requirements document (PRD), handing work to designers and engineers, and managing a release schedule. A document is a tool for building a product. First, the team needs to decide which problem it is trying to solve.
Before improving a sign-up flow, for example, you need to know where customers leave. Are there too many fields? Is something going wrong during verification? If the numbers alone don’t explain why, customer inquiries and interviews can help.
The PM’s role is to connect scattered information and clarify the problem for the team. Customer support knows which questions keep coming up. Data specialists can show where users drop off. Designers and engineers can investigate usability and technical causes.
AI can organize a large amount of information quickly, but the words that appear most often in support requests do not always identify the problem most worth solving. The team still needs to judge which customers are affected and how much the issue matters to both the customer experience and the business.
PMs can build more for themselves

Once the problem is clear, the team can discuss solutions. One approach is to write a plan, have a designer create the screens, and then ask an engineer to begin development.
Now a team can also try a prototype before the document is complete. A PM might use AI to create an alternative sign-up flow and show it to colleagues. Seeing a version of the experience makes it easier to discuss where customers get stuck and whether the proposed screen could help.
Mark Kwon, a PM at Mercari, shared his experience of working on and releasing small features himself in a team that already had a design system. A design system brings together reusable interface elements—such as buttons and input fields—and the standards for using them. With that foundation in place, a team can use existing components to build simple features more quickly.
Still, being able to create a prototype is different from being responsible for a product in operation. Even if AI creates a screen, someone needs to check whether customer information is handled safely, whether the feature conflicts with existing functionality, and whether the team can recover from a failure. Features affecting payments or personal data need thorough review by the relevant specialists.
When PMs can make more for themselves, they can see an idea earlier and have a more concrete discussion with designers and engineers about what still needs review.
What should a PM own in the AI era?

A PM’s role and responsibilities vary from company to company. Some own a broad product strategy; others focus on a particular feature or customer experience. Rather than relying on job titles alone, teams should agree on who makes which decisions at each stage.
The following is a useful starting point for dividing responsibilities:
| Stage | What the PM typically owns | Work shared with others |
|---|---|---|
| Discover the problem | Define the customer problem and explain why it needs solving. | Review support requests, data analysis, and customer interviews together. |
| Set priorities | Recommend what to tackle first, based on customer impact and business goals. | Leaders and relevant teams review costs, timing, and business direction. |
| Shape the solution | Explain the problem and intended outcome; create a prototype when useful. | Designers review the user experience; engineers review implementation and constraints. |
| Build and check quality | Define the release scope and completion criteria; review the result from the customer’s perspective. | Engineers and quality assurance (QA) specialists handle technical quality and testing. |
| Learn after release | Review customer responses and metrics; decide whether improvements or further experiments are needed. | Marketing and customer support share feedback from the market and customers. |
AI can help organize information, draft material, and reduce repetitive work at each stage. But the team must decide what action to take and who is accountable for it.
Suppose AI analyzes thousands of support requests and concludes that the sign-up screen is confusing. Before turning that conclusion into a requirement, the PM should examine the evidence: Where do users actually drop off? Do the customers who contacted support reflect the experience of other users? The answers help the team design a more meaningful experiment.
Blurred boundaries make expertise more important

As AI makes it easier for PMs to try design or development, engineers and designers can also participate more readily in customer research and product planning. The shift is toward people contributing across a wider range while bringing their own expertise.
An engineer might understand a customer frustration and suggest a small improvement. A designer might use AI to build a prototype and show it to users sooner. A PM might explore data or a screen before requesting development.
Working this way requires understanding other disciplines well enough to evaluate the results. Even if AI writes marketing copy, it is difficult to judge the draft without knowing who it is for and which channel will carry it. If AI writes code, someone unfamiliar with personal-data handling or possible errors may not know when to ask for a review.
PMs do not need to master every detail of development, design, and marketing. They do need to learn the basics and distinguish what they can check themselves from what requires a specialist. Using AI well means not only getting an output quickly, but also being able to question and review it.
Fast experiments and careful releases can coexist

If AI reduces the time needed to build a prototype, a team can test more hypotheses. Changing sign-up button copy or trying a different layout for a confusing screen can begin on a relatively small scale.
A feature that affects payments or personal data is harder to rush through in the same way. The harm from a failure may be greater, and reversing the change may be difficult. Sakana AI’s careful management of user access to its cybersecurity-related model is one example of adjusting the approach to the risk.
When proposing an experiment, a PM should consider four questions alongside speed:
- Impact: Which customers could be affected if something goes wrong, and how?
- Test scope: Can the team start with a limited group or an internal environment?
- Recovery: Can the change be reversed if the result is poor?
- Review: Who will check the AI output and the quality of the actual release?
Small changes can be tested quickly, while higher-impact decisions receive more scrutiny. A PM must judge which work can move quickly and how quickly it is appropriate to move.
The questions an AI-capable PM needs to ask

As AI writes documents and builds prototypes, the questions a PM asks at the beginning become even more important. If the problem is unclear, a quickly produced solution can still take the team in the wrong direction.
Before starting a feature, try writing down the answers to these questions:
- Which customer and which problem is this product intended to serve?
- What evidence tells us the problem matters now?
- How will we know whether our proposed solution works?
- What result would make us continue, and what would make us reconsider?
Customer feedback and data may point in different directions. Team members may propose different solutions. A PM helps surface those differences, discuss them with the right people, and test the ideas on a small scale.
AI helps PMs research, build, and learn faster. It also lets them try more work themselves. That makes it increasingly important to decide not only what can be built, but what should be built and who will be accountable for the outcome.
In your team, which customer problem does the PM own—and do they have the authority to follow through on the outcome? Is the time saved by AI going toward building more features, or toward finding more important problems to solve?
Read Next
More posts related to AI Trends →
How Do You Build and Connect AI Agents? Notes from Google Agentic AI Night
How do you build AI agents and connect them so they can take on a task together? Start by defining each agent's role, the instructions it should…

Teach Kids About AI, Limit Their AI Use: 6 Family Rules for Safe AI
In our last post, we looked at "What Should Kids Learn When AI Does the Work?" This time, let's look at what families can actually do at home. O…

AIB Featured by Levtech Freelance: What Will Engineers Bring to the AI Era?
We are pleased to share that AIB has been featured in Levtech Freelance’s guide to AI tools and agents for engineering productivity, published f…

What Should an AI Agent Remember? How 'Dreaming' Fixes Its Memory
Give an AI a "memory notebook," and it gets smarter on its own, right? Not quite. In research Anthropic published on November 6, 2026, notes alo…