How AI is Redefining Product Management
A task-by-task analysis
In a world where everyone has their own different view of how AI is going to change things, many people end up operating in one of the two modes. In one mode, they become hyper-vigilant tracking every information that comes through newsletters and professional feeds. In the second mode, they end up ignoring everything because nothing seems certain and there is simply too much noise.
Both of these modes are extreme and counterproductive. What one needs is measured, selective engagement with AI by focusing on the developments relevant to our work.
The next question becomes: how do we adopt such measured approach for our professions? This post is written with an aim to answer this question for product management professionals.
Answering this question is difficult because of two reasons. First, AI is a rapidly emerging field. Every week, we are hearing about novel developments. So even if we can create a list of things that AI can do/help today, it won’t be valid probably next quarter. Second, the context of the organisation matters when using AI professionally. This is because each organisation has a different level of reception of AI. Add to that, the effectiveness of AI is uneven across topics and domains.
We would like to focus on solving the first problem — how do we create and maintain the list of things AI can help with? Once we solve the first problem, I have confidence that PMs can apply it in their organisation because they understand those complexities/hurdles well.
To create a list, we will start with a simple task classification framework. This framework can help us assess how to look at tasks in context of AI. As a step 2, we can take most of the tasks that PMs are supposed to do, and show how we apply this framework.
While the list of tasks will be long, going through them one-by-one will give you the confidence to apply it yourself. One clue that you have become comfortable in applying the framework is if you start getting bored towards the end. That means, you are able to predict what I am going to write for the next task and the task after that :)
Lead-Edit Framework
One of the concerns that PMs have around AI is whether it is going to replace them. It’s a legit concern, but doesn’t take us anywhere since we can’t predict the future. What we can do today is focus on how AI can enhance PM skills by performing tasks relevant to a skill. To know which ones it enhances, doing a task by task analysis is useful.
A simple classification framework that I would propose is Lead-Edit Framework. All tasks can be classified based on whether the PM has to lead the task (and AI can assist), or edit the task mostly done by AI.
There are four categories in this lead-edit framework:
PM-led tasks (AI-assisted): These are the tasks where AI can help PMs become more productive. The primary control lies with humans. For example, to create a product roadmap, we need to combine the org knowledge, market research, and product context. While AI can assist in market research, competitor research, etc, it can’t bring various context together to create a roadmap. So roadmapping is a PM-led, AI-assisted task.
PM-edited tasks (AI-led): These are the tasks where AI can lead the way, but would require humans to do the final validation. PMs work as an editor in these tasks. For example, if we provide the goals of the user research to AI today, it can create a comprehensive research script. PMs need to review and refine it. See this sample script generated for Flipkart Minutes to Claude Sonnet 3.5 when we gave it this prompt. Adding part of the script below for you to see how good it gets!
Sample Research Script PM only tasks: These are the tasks that only PMs can do, such as resolving a conflict between team members.
AI only tasks: These are the tasks that AI can do on its own, like generating a daily status report of all tasks based on project management tool.
Classifying Tasks
To classify tasks, we need to create a list of tasks. We can divide the tasks into two parts — one is the set of tasks expected from Junior PMs, and the other is the set of tasks expected at senior levels. At junior levels, you are expected to research, ship, and iterate features/ products. At senior levels, you are expected to manage a product or suite of products, their strategy, roadmap, and growth. As you move up further, you are expected to hire and build an org + contribute to company strategy.
If you confused by what Junior vs Senior implies, below is a good set of skills and their bifurcation to help you.
The tasks, by no means, cover everything but they do a reasonable job of covering 80% of what you are expected at these roles.
The classification will help you in 2 ways:
It helps in identifying opportunities to be more productive as PMs through AI assistance
It highlights which PM skills become redundant, which ones stay relevant and which ones require deeper knowledge.
List of Tasks
Classification of Tasks in Junior Roles
The Tools I used to perform tasks and gauge AI effectiveness are mostly chat tools and APIs from leading foundation models: Claude Sonnet 3.5, ChatGPT-4o, Claude Developer console. Agents automate tasks, but the ability/ skill level can be seen through the chat interfaces only! In addition to doing these tasks by myself, I also relied on industry papers around what sort of tasks can LLMs execute and to what efficiency.
Here are few generalisations that we have seen so far:
Code-related tasks like writing queries, executing queries, etc. can be increasingly handled by AI only. We have tools where you can write natural language queries such as ‘show me sales volume in USD from Jan’25’ and they can write/execute queries on your behalf to bring you numbers.
Social interactions and relationship management can be handled by Human only
Complex decisions like roadmapping requires you to utilise knowledge from multiple areas - data, business, marketing, etc. While AI can assist humans in this, such tasks right now are human-led.
Summarisation, documentation and other activities where the stakes are lower can be handled by AI for the most work. They just need humans to validate the output. This is where PMs would work like an editor. Such tasks will be classified into AI-led, PM-edited.
In a future post, I will cover how to use prompt engineering to be more effective at understanding how well can AI assist PMs. But for this post, let’s stick to looking at these tasks and classifying them. If you don’t understand/agree with few of these tasks classified as such, feel free to ask questions in comments.
Discovery (user research)
Creating a research goal and plan: The goals of a research must align with business or user problems. Setting a research goal is high-stakes and also complex problem, that’s why it has to be PM-led, AI-assisted.
Let’s talk a little about how AI can assist you. AI can be helpful if you write good prompts sharing context. The output of AI can improve further if we add business and product documents for context. But more of this in a future post around prompt engineering. we can’t fully rely on it, since a mistake at this level can be costly!
Creating a research script/ survey: AI-led since once we have provided goals and plan to AI, AI is pretty good at generating script with questions. PMs need to act as an editor here.
Recruiting users: Recruiting users for research requires reaching out to TG, filtering the right users for research, and defining incentives for them. For offline recruitment, this task is PM-led. For online recruitment, AI can do a reasonable job but PM judgment might be needed for final selection.
Conducting interviews: The moderator needs to improvise in real-time often to gain insights. The conversation also needs to feel natural. AI can act as a note-taker here. While AI can conduct interviews even now, we aren’t certain it can repeatedly do this job well without hallucinating. For this reason, conducting interviews online is PM-led. Conducting interviews offline is PM-only.
Generating insights from interviews and reporting: AI is pretty good at going through numerous transcripts and filtering novel, important insights. AI also excels at pattern recognition in interview data, so it can note down the patterns that occur more frequently. This is why summarisation, insight generation, and reporting can be done by AI. PMs need to act as editor here.
PMs need to lead overall research process but it can be significantly enhanced by AI. In the follow up tasks, I would not elaborate much in task classification unless necessary because by now you would have gotten a good sense of how this classification works.
Shipping a feature/product
Shipping a feature requires orchestrating various moving parts.
Exploring solutions to a problem: AI can generate solutions, but evaluating it needs human judgment. A lot of times, the solutions generated by AI are suboptimal, so PMs still need to devote a significant time coming up with the solution. Since this is a high-stakes situation and we are also not sure about AI’s output, this would be PM-led, AI-assisted.
Writing PRD: With all the context of discovery, problem, and solutions, writing a PRD can be a AI-led activity. AI can draft detailed PRDs, PMs can act as the editor.
Convincing different stakeholders: Human only because we need to manage relationships, understand motivation of various stakeholders and manage them. Though at times, I have personally found AI advice to be pretty useful when stuck around how to handle a particular scenario.
Prioritising feature: PM-led, AI-assisted since it is a high-stakes activity and AI might not have the full context of feature reach, impact, effort, strategic goals, etc. That said, I do see AI making strong progress in prioritisation as it gains better context.
Early prototyping: Tools like Replit, Cursor, Lovable, Bolt, etc. help you create quick high-fidelity, early prototypes. These early prototypes can help PMs communicate the requirements well with designers and other teams. Early prototyping is an AI-led activity, and one that’s becoming increasingly important for PMs to learn for efficiency.
Sprint planning: PM-led activity since it requires understanding team capacity, dependencies, prioritisation, etc.
Tracking the status of the task: AI-only since it is a pure data tracking task.
Defining success/ failure of the feature: AI can do it reasonably well today for products similar to popular products (FAANG portfolio). But it can’t do well where you are building something novel. For that reason, I am going to classify it as PM-led.
Tracking the metrics: AI only
Sending a status update after launch: If success criteria is defined well, a status update can be sent by AI only.
Partying with the team on a successful launch: Human only.
Conflict resolution: Human only as it requires emotional intelligence and relationship management. That said, AI can be a good listener and help you figure ways to resolve conflict.
Iterating the feature
Finding areas to improve upon using data and user research: PM-led, AI-assisted
If we are being honest, most of the work that Junior PMs are doing can benefit heavily from AI, and those who make use of it would get a lot done. It kind of feels like we are binning to see the birth of 10X PMs.
Let’s see how AI redefines senior product roles.
Classification of Tasks : Senior Roles
Product strategy and growth
Product vision and long-term goals: Setting up product vision and long-term goals is a human-only activity today since humans can retain context over years and can assimilate the context of everything around them - economic, social, legal, ethical, moral, etc.
Competitor analysis: AI can gather, analyse, and track competitor movements. PMs need to systematically analyse what AI throws at them, and make decisions. Competitor analysis is AI-led for this reason.
User needs and preference: Understanding needs and preferences of other humans requires empathy. AI can process research data and identify patterns, but understanding deeper needs and behaviours requires human insight. Such human insight also becomes a key competitive advantage. That said, many PMs I meet haven’t spent enough time in human psychology and behavioural economics. It’s high time they should start investing in these areas.
Critical and creative problem solving: Competitor research and user research point towards the business or user problems that need to be solved. PMs are way better than AI in this particular aspect. AI can be useful in brainstorming and can bring case studies to inspire some solutions here.
Creating roadmap: Roadmap requires prioritisation and alignment with overall org strategy + product vision. AI can help with useful data for prioritsation, but can't lead this strategic exercise. Roadmap, for near future, is going to be a PM-led activity.
Finding GTM/channels to grow the product: Product growth is more of a science which utilises data around budget, CTRs, conversions, etc. Finding channels to grow the product is something AI can do really well, and just needs PMs/ marketers as editors. Where PMs need to focus on is novel areas such as product-led growth.
Product-led growth: Product-led growth is fundamentally merging product development, strategy and growth.
Pricing: AI-led since AI can analyze market data, competitor pricing, and suggest optimal price points. It can also use some pricing research tools. PMs can act as an editor and orchestrator here since it is a quite collaborative activity.
Positioning: Product positioning requires deep understanding of user psychology and brand values. AI can assist with competitor analysis and market data.
Leadership
Setting goals (OKRs): PM only
Adjusting goals: PM-led, AI-assisted. AI can help in tracking goals and suggest adjustments based on historical data and patterns in current data.
Financial planning and resource allocation: PM-led, AI-assisted because financial planning and resource allocation requires complex decision making.
Cross-functional alignment: PM only
Technical strategy: PM-led, AI-assisted since AI can provide case studies/ how other companies have done their technical strategy.
Hiring and building a strong team
Interviewing and hiring PMs: PM only since it requires nuanced evaluation in hard skills, soft skills, and cultural fit.
Onboarding PMs: AI can help with good onboarding by managing the onboarding process with right documents. Human mentorship is crucial for PMs to be effective in the early stages. So we will classify it as AI-led.
Performance management: Managing performance requires human judgment, empathy and mentorship. A lot of conversations around performance management happen offline and require deeper relationships between the manager and the team.
Corp strategy
Pitching for new initiatives
Pitching for investment in existing product lines
Product portfolio management
Build vs buy vs partner decisions
All of these tasks are PM-led or PM-only. AI can help in market analysis, brainstorming, etc.
What can PMs do?
If you look through all the tasks that PMs do currently, you will realise two things.
First, AI can easily make a PM 2-3x effective today. PMs need to start using AI as a productivity multiplier. But it shouldn’t be limited to just writing simple prompts and asking ChatGPT or Claude questions that come to their mind. They have to build new systems and processes for themselves that make them more effective. This includes, but is not limited to,
Writing effective prompts for AI tools
Use it in documentation, analysis, research
Automate routine tracking and monitoring
Create explorations in design
For product leadership, this means helping the team setup these systems and processes. The leadership also has to start pushing teams in the areas where PMs can add lots of value in the AI edge. This could mean creating new guidelines for AI usage, performance management, etc. If you are not doing it, your competitors are and that becomes their advantage.
The second realisation is that AI is increasingly doing a lot of tasks better. What it means for PMs is that they have to focus on strengthening the Core Competencies that AI can't replace. A proxy for that are areas which are inherent weakness of AI
Product strategy and Corp strategy
Stakeholder management and relationship building
Critical problem-solving
Creative thinking
Team leadership
Conflict resolution
User empathy
..
As AI progresses, I will keep re-doing this exercise of re-evaluating tasks. Do subscribe to stay updated!
Prototyping with AI is a slightly complex topic because it needs you to use prompt engineering, technical understanding, and design sense together. At the same time, it is quite useful for PMs. I would be doing a live 2-hour workshop (open to all) on this topic on 6th February, 9 pm - 11 pm IST. You can register for the free workshop here
This would be all for this edition!
Thanks,
Deepak