Can AI replace a Product Owner? What still needs a PO

A balanced answer for new POs: AI helps with drafting, summarising and suggesting, while accountability for product value and the decisions behind it stays with a person.

Written by Phan Lam Duy Khoi

In this article's view, AI does not replace a Product Owner, because the Scrum Guide 2020 makes one person accountable for maximising the product's value, and that accountability does not move to a tool. AI helps with drafting, summarising, organising and suggesting options; framing the problem, checking evidence, making trade-offs between people and owning the outcome remain the PO's work.

A hard question in your first week

This is an illustrative situation. You have just become Product Owner (PO) for Bếp Nhà, an app for ordering cooking ingredients (a fictional name). In a meeting, your manager, who has just read a news article about AI, asks half jokingly: “AI tools can write user stories now, summarise customer feedback, even put together a roadmap. Does our team still need a PO?”

You are not sure what to say. That evening you scroll social media. Half the posts say AI will replace product management altogether; the other half say AI is a toy and should not be trusted. You are worried about your new job, and you feel you should be using AI, but you do not know how far to take it.

This article does not predict how many jobs will be lost or kept. It gives you a way to answer the question for your own work: separate the drafting work from the deciding work, then see which parts AI can reach and which still need a person who is accountable.

Why is “can AI replace a PO” hard to answer?

First, what AI produces looks a lot like what a PO produces: a user story, a feedback summary, a priority list. If you judge the PO role by its documents, it is easy to conclude AI already does most of it. But documents are only the record. What matters is who chose to write that, based on what evidence, and who answers for it when it is wrong.

Second, the Scrum Guide 2020 does not define the PO by a list of documents. It says the PO is accountable for maximising the value of the product resulting from the work of the Scrum Team, and stresses that the PO is one person, not a committee. The PO is also accountable for effective Product Backlog management: developing and communicating the Product Goal, creating and communicating backlog items, ordering them, and keeping the backlog transparent. Scrum.org's page on the PO role also describes the PO as the person who balances the needs of many stakeholders and makes the final call on the backlog after hearing their input.

The Scrum Guide also says the PO may do this work or delegate it to others, but either way remains accountable. In this article's view, handing drafting to an AI tool works the same way: the task can move, the accountability does not. A tool does not stand in front of stakeholders to explain why one feature goes before another.

Third, the anxiety pulls people in two directions. NIST, the US National Institute of Standards and Technology, describes both in its generative AI risk profile. One direction is people being unnecessarily averse to AI and missing out on its benefits. The other is “automation bias”: deferring too much to automated systems and seeing AI content as better than it is. NIST also lists “confabulation”, confidently stated but false content, among the main risks. New POs can easily fall to one side, because they do not yet have the experience to see where the line is.

How to answer it yourself: separate drafting from deciding

Instead of asking “can AI replace a PO” in general, ask something more specific: this week, which of my tasks are drafting and which are deciding? Drafting turns information you already have into words, tables and lists. Deciding means choosing what is true, what goes first, what to tell whom, and owning that choice.

Four steps to see where AI reaches in your work Do it once with a real week of your calendar. Steps 3 and 4 matter more than steps 1 and 2.
  1. List the work: Write down a real week, including meetings and messages
  2. Label it: Each task is drafting, deciding, or both
  3. Drafting part: AI can help, then check every point against real sources
  4. Deciding part: Keep it, do it more carefully, record the reason

If you cannot tell how to label a task, ask: if the result is wrong, who has to explain it? If it is you, the task has a deciding part.

Much of a PO's work has both parts. Writing a user story has a drafting part (wording, structure) and a deciding part (which user, which business rule, how far the scope goes). Separating them helps you use AI in the right places without letting it creep into the places that need you.

For the drafting part, the rules for checking an AI draft are in the separate article on using AI in PO work. For data, read the article on what not to put into AI tools before you paste anything from work.

Which PO tasks can AI do, and which still need a PO?

The table walks through activities tied to the PO role in the Scrum Guide and in daily work. This is the article's view, not an official list from the Scrum Guide.

PO activities: where AI helps and what stays with the PO
PO activityAI can help withStays with the PO
Set and communicate the Product GoalPhrasing ideas, drafting the presentationChoosing the goal, weighing it against company strategy, getting others to commit
Order the backlogListing comparison criteria, summarising each itemDeciding the order, giving the reasons, answering for it if it turns out wrong
Learn about usersSuggesting interview questions, grouping notes cleaned of personal detailsMeeting real users, hearing what they do not say, telling opinions from behaviour
Clarify work with DevelopersDrafting user stories and acceptance criteriaAnswering the team's questions, settling business rules, choosing what to leave for later
Work with stakeholdersDrafting messages, summarising meetingsTrading off between groups, saying no, keeping promises already made
Measure results after releaseSuggesting questions to ask of the dataChecking the data source and the calculation, deciding to continue, fix or stop

Developers here is the Scrum Guide's term for the people in the Scrum Team who build the product, not only those who write code. What the entries under “Stays with the PO” have in common is that other people act on your answer. When Developers ask “this part first or that one?”, they need a person who answers and stands behind the answer.

The Scrum Guide says that for the PO to succeed, the entire organisation must respect their decisions. That respect attaches to a named person with context and a track record of keeping their word. A tool can offer a plausible option, but it has no track record of keeping its word with anyone.

How might the PO role change with AI?

This section is the article's view, not a forecast backed by data. As drafting gets faster, a PO's time may shift towards three things: framing the problem clearly before drafting, checking more carefully after drafting, and making more decisions with clearer reasons. In other words, less of a PO's value lies in writing the document, and more in knowing whether the document is right and whether the work is worth doing.

Four skills worth investing in The four support each other. You can pick one to practise first this month.
  • Framing the problem: Say in one sentence who is struggling, where, and what changes once it is solved. AI drafts better when the problem is clear, and checking gets easier.
  • Checking evidence: For every number, rule and conclusion, ask: where is the source, who confirmed it? Go back to the original instead of trusting the summary.
  • Deciding with reasons: Choose between options, state the cost of each, record why you chose. These reasons are what the team and stakeholders need to hear.
  • Keeping people's trust: Understand each group's context, say no without losing goodwill, keep your promises. Trust is built over many rounds of real work.

None of these skills is new. They were the hard part of the PO role long before AI tools. What changes is that when drafting becomes cheap, these skills become more visible: two POs using the same tool get very different results if one can frame and check while the other only copies the draft.

Example: a week of work, where does AI touch it?

Back to the PO at Bếp Nhà. The table walks through an illustrative working week, from the first day to the last. This PO's company has a policy allowing one AI tool for documents that contain no personal data. Each row shows what AI helped with and what it could not do.

A week for the Bếp Nhà app's PO (illustrative example)
DayMain taskWhat AI helped withWhat AI could not do
Day 1Read customer feedback on late deliveriesGrouped feedback, cleaned of personal details (names, phone numbers, order IDs, addresses), into five themesRereading the originals for the largest theme showed it was really two different problems
Day 2Clarify the request “let customers pick a delivery slot”Drafted the user story and acceptance criteriaAsked operations for the real number of slots and the order cut-off time, corrected the number the AI had filled in
Day 3Order the backlog for the next sprintListed criteria for comparing itemsChose delivery slots over the e-wallet feature and explained why to the sales team
Day 4Meeting with sales and operationsSummarised the meeting notes into a task listNoticed the two teams meant different things by “fast delivery” and agreed one definition before sending the notes
Day 5Sprint Review and weekly numbersSuggested questions to ask of the dashboardChecked how the late-delivery rate was calculated, decided to keep working on slots instead of expanding to a new area

Across the week, AI appeared on all five days and saved the PO time on drafting. But each day, the most important result came from “What AI could not do”: a feedback group merged wrongly, a number filled in wrongly, a word two teams understood differently, a choice that needed someone to stand up and explain it.

The example follows the method above. Every task has a drafting part handed to AI and a deciding part kept by the PO. The data given to the tool was cleaned of personal details, not just names but order IDs and addresses too, and stayed within what the company allows. If your company has no such policy yet, ask before you use one.

Self-assessment: which of my tasks are drafting and which are deciding?

Use this checklist with a real week of your work. The point is not to score yourself but to see clearly where your time goes and which parts need more care.

Checklist: drafting or deciding
Week: ...........

Step 1. List the 8 to 10 main tasks of the week (including meetings, messages, documents).

Step 2. For each task, answer:
[ ] Is this mainly turning information I already have into words, tables or lists? (drafting)
[ ] Does this need me to choose between options or settle what is true? (deciding)
[ ] If the result is wrong, am I the one who explains it to the team or stakeholders?
[ ] Does this need context only people inside the company know (policies, history, relationships between groups)?
[ ] Does this need a direct conversation with someone?

Step 3. For the drafting part:
[ ] Does the company allow AI tools for this type of document?
[ ] Does it contain personal data, contracts or internal figures? If so, keep it out until it is allowed.
[ ] Do I already have confirmed facts before asking AI to draft?
[ ] Which source will I check each number, rule and conclusion against?

Step 4. For the deciding part:
[ ] I can state the reason for this decision in two sentences.
[ ] I can state the cost of the option I did not choose.
[ ] I have recorded the decision and the reason where the team can see it.

Step 5. Looking back at the week:
- Which tasks am I drafting myself that AI could draft first?
- Which decisions am I rushing that need more time or evidence?
- Which of the four skills (framing, checking, deciding with reasons, keeping trust) do I want to practise first?

The question “if the result is wrong, am I the one who explains it?” is the easiest way to sort tasks. If the answer is yes, the task has a deciding part, even if it looks like simple writing.

Mistakes new POs make when thinking about AI and the PO role

  • Avoiding AI entirely out of fear or because using it feels like cheating. NIST also notes that unnecessary aversion makes people miss the benefits. If your company allows it, try it on drafting work and learn how to check it.
  • Letting AI decide. Taking an AI priority order or conclusion as the final answer without checking the reasons. This is the excessive deference NIST calls automation bias.
  • Judging the PO role by documents produced. Writing more user stories does not necessarily create more value. The Scrum Guide ties the PO's accountability to product value, not to document count.
  • Putting sensitive data into a tool to “save time”. Customer data, contracts and internal figures need the company's permission first. Ask your legal or IT team if you are unsure.
  • Worrying about your career based on headlines. A news article does not know what your job involves. The self-assessment above gives you an answer closer to reality.
  • Presenting AI output as your own judgment without checking it. If a stakeholder asks “why?” and you cannot answer, you lose trust faster than the time you saved.

Summary

  • According to the Scrum Guide 2020, the PO is accountable for maximising product value and is one person, not a committee. That accountability does not move to an AI tool.
  • AI helps with drafting: document drafts, summaries, grouping, listing options. The PO keeps the deciding: framing, checking, trade-offs, accountability.
  • Answer “can AI replace me” by listing a real week of work and labelling each task as drafting or deciding.
  • In this article's view, the PO role may shift towards framing, checking and deciding. Four skills worth investing in: framing problems, checking evidence, deciding with reasons, keeping trust.
  • Avoid both extremes: shunning AI or trusting it too much. And always check your company's policy before putting data into a tool.

I am writing more PO situations in this same format, with templates you can use straight away. If you would like the next one by email, subscribe to the newsletter below.

Frequently asked questions

Can AI replace a Product Owner?

In this article's view, no, because the Scrum Guide 2020 makes one specific person accountable for maximising product value. AI speeds up drafting and summarising, but decisions, trade-offs and responsibility for results stay with the PO.

What should a Product Owner use AI for?

Use it for drafting work: user story drafts, meeting note summaries, grouping feedback, listing options. Then check every point against real sources before using it, and keep sensitive data out of AI tools unless your company allows it.

What skills should a PO build now that AI exists?

Four are worth investing in: framing problems clearly, checking evidence, making decisions with explained trade-offs, and keeping the trust of the team and stakeholders. In this article's view, these are the parts an AI tool cannot do for you.

Sources to check

Related articles