The AI Product Manager

Author:Murphy | View: 21030 | Time: 2025-03-22 18:47:58

When ChatGPT was launched at the end of 2022, it started a new chapter in AI. Since then, AI has grown quickly. Almost every week, we hear about new, more powerful models. Millions of people worldwide use these tools daily, and companies are trying to understand how to take advantage of this incredible technology. I say "trying" because although some fields have managed to find great applications for this (coding copilots, marketing, or customer support), many others are still discovering, prototyping, and understanding how to use it and how to deal with the new challenges and risks that come with it (more on this at "2025: AI Eats the World").

So how can companies close the gap between these powerful models and the potential business value? This is where the AI Product Manager (AI PM) comes in.

An AI PM is responsible for building AI-powered products that solve user problems, and bring business value, while building trust and dealing with the risks of AI.

An AI PM differs from a traditional PM role mainly on the type of solutions their team builds: strong focus on ML & AI models, importance of data for training and prediction purposes, handling unpredictable outcomes (AI is probabilistic, not deterministic like traditional software), and making sure AI is used responsibly and users understand and trust the team's solutions.

In this post, I'll introduce the role of an AI PM according to my experience working in this role for the last 3 years. We'll explore the different types of AI PMs, their main skills, tasks and daily responsibilities, and what are the biggest challenges to face in this position.

Types of AI Product Managers

As with many roles in tech, the definition of an AI PM can vary widely depending on the company. However, from my experience, AI PM roles typically fall into four main categories, each with distinct responsibilities and focus areas:

AI Research PMs: work in teams focused on developing foundational models, general-purpose and business agnostic models, or research oriented AI solutions. These solutions are often offered as services to other companies, as seen with organizations like OpenAI or Anthropic, where the models themselves are the primary product. These teams typically consist of Data Scientists with a strong research background (often called Research Scientists) and may include other roles like Machine Learning Engineers to support infra, model deployment and scaling.
AI Platform PMs: work with teams that build internal AI or ML platforms. These platforms enable other teams within the company to train, deploy, and maintain their own ML models or AI solutions. In this case, the product is the platform, and the team's focus is on solving problems and delivering value to the internal users who need to train and deploy AI solutions. Teams usually will include Machine Learning Engineers and other technical roles such as Platform Engineers, to develop and maintain the platform.
AI core PMs: work in teams that act as internal consulting for other product teams requiring specific ML or AI models. These models are usually integrated into broader solutions by the requesting product team. Since the strategy and requirements for the solution are defined externally, the AI Core PM role often focuses more on delivery than discovery (more on this later). These teams primarily consist of Data Scientists who build the models, while the consuming team will have the roles needed to integrate the solution into their features or functionalities.
AI product PMs: focus on applied AI, integrating AI capabilities into specific products that impact end users and the business. As a result, the team includes not only Data Scientists and Machine Learning Engineers, but also Backend Engineers, Frontend Engineers, and even iOS/Android Engineers.

In this post, I'll focus on the Applied AI PM role, although this role has a lot of things in common with the other three types of AI PMs. Applied AI PM is the role I currently hold and love: working with multidisciplinary teams to deliver AI-powered features and functionalities that bring value for users and businesses. To learn more about how a team like that operates, check out my previous post:

Working in a multidisciplinary Machine Learning team to bring value to our users

AI PMs main skills

The four key skill sets for an ML / AI PM, image by author

There are many necessary skills and knowledge needed to succeed as an ML / AI PM, but according to my experience, the most important ones can be divided into 4 groups:

Product strategy: understand users and their pains, identify the right problems and opportunities, prioritize them based on quantitative and qualitative evidence, and aim to impact product and business metrics.
Product delivery: manage a team's initiative to deliver value to the users efficiently, ensure clarity on what to build, and unblock the team and manage dependencies.
Influencing: gain trust, align with stakeholders and guide the team.
Tech fluency: knowledge and sensibility in Machine Learning, Responsible AI, Data in general, MLOPs, and Back End Engineering. Ability to act as a bridge between the product opportunities and the AI solution possibilities.

For a deeper dive into each of these topics and relevant resources to learn each of them, you can check out my previous post:

From Data Scientist to ML / AI Product Manager

AI PMs main tasks and daily responsabilities

Like any other PM, an AI PM is responsible for setting the strategy and overseeing the delivery of the team's work, but the team doesn't report directly to them. This means they don't have to handle people management tasks – that's typically the responsibility of an Engineering Manager or Data Manager. However, they still need to guide the team's direction through influence rather than authority: by building trust, fostering collaboration, and creating a culture of empowerment.

What best defines my day-to-day as an AI PM is balancing continuous delivery with continuous discovery during each sprint (typically two weeks). This balance is essential to ensure the team consistently delivers value while staying aligned with the company's goals.

Discovery and delivery tracks, image by author

Continuous delivery tasks

On one hand, an AI PM needs to make sure the team successfully implements solutions, iterations, and key tasks throughout the sprint. Continuous delivery ensures they provide value to users and the business continuously. The main tasks the PM needs to do to achieve this are:

Managing agile rituals: Leading daily stand-ups, sprint planning, and backlog refinement.
Defining clear tasks: Ensuring tasks are well-defined, actionable, and aligned with sprint goals and product strategy.
Handling dependencies: Proactively managing dependencies and anticipating blockers to keep the team moving forward.
Balancing workloads: Distributing and planning work effectively across roles while empathizing with individual strengths and weaknesses.
Effective communication: Speaking the "language" of all team members to engage meaningfully, challenge ideas, and provide feedback.
Evaluating use cases: Helping define the right evaluation metrics for each specific use case.
Starting small: Challenging the team to start with Minimum Viable Products (MVPs) and helping define it, instead of attempting huge projects from the start.
Measuring progress: Ensuring progress is measurable and visible to all stakeholders.
Boosting morale: Acting as the team's cheerleader by fostering optimism, celebrating wins, and keeping the team motivated.

Continuous discovery

On the other hand, the AI PM ensures the team discovers the right problems to solve and identifies what to build next. Key aspects of this work from the PM side include:

Defining vision and strategy: Creating and maintaining a clear product vision, strategy, and roadmap that prioritize feasible, impactful, and ethical AI solutions.
Staying informed: Keeping up-to-date with business strategy, AI advancements, and industry trends.
Analyzing competitors: Conducting competitor analyses and exploring AI use cases to inspire and guide the team.
Gathering user feedback: Collaborating closely with UX and data teams to collect qualitative and quantitative insights from users.
Facilitating collaboration and discovery: Organizing workshops with the team to share relevant insights, brainstorm ideas, and prioritize next steps.
Connecting metrics to outcomes: Translating technical model performance metrics (e.g., accuracy) into measurable product or business outcomes (e.g., CTR or user satisfaction).
Promoting AI literacy across the company: to align expectations and ensure a shared understanding of what AI can and cannot achieve.

Continuous delivery and continuous discovery might sound simple, but as you can see, they translate into a lot of critical responsabilities. Managing to find the right balance between the two is key. Too much focus on delivery means the team will eventually run out of meaningful tasks or next steps, and that the solutions built might miss the greatest opportunities to deliver user value. Too much focus on discovery results in great ideas with little actionability and limited learning from real-world feedback.

On top of balancing delivery and discovery, there are always smaller tasks that come up during the sprint: preparing summaries for upper management, meetings with other teams, addressing user or stakeholders requests… To handle these effectively, it's important to adopt a hands-on approach by being proactive (e.g. learn to run queries and uncover insights on your own) and resourceful (e.g. gain the technical depth needed to solve smaller team requests independently). Having this autonomy allows you to handle smaller requests independently, unblock the team when needed, and minimize distractions so they can stay focused on high-priority work.

What makes the job hard

Although the AI PM job is a great job that can be a lot of fun and with great opportunities and career growth, there are some big challenges that can make the job hard according to my experience:

Pressure to know what's next: you are continuously expected to provide clarity and direction, even when the future feels uncertain. On top of this, the AI field is advancing so fast (new model capabilities, functionalities, costs and latency changing fast…) that makes knowing what's next really complicated!
Time management and multitasking: switching between sprint planning, user research, presentations, and meetings demands constant focus-shifting. This can be exhausting!
Creating your own playbook: AI PM is still new and usually there won't be many in the company, so you need to establish your own processes since traditional PM frameworks often don't apply.
Black-box frustration: AI models, especially large language models, often function as black boxes. Debugging or defining next iterations to improve them can feel challenging.
Lack of Control Over Probabilistic Systems: AI's probabilistic nature requires rethinking product design, testing, and user expectations. Guiding your team through this shift is part of the role, and also a challenge of constant trade offs: "is this model performance or this evaluation results good enough to go to production?"
Ethical challenges: AI ethics is a very complex topic, where risks can come from many different sources, and fighting them is not straightforward. Even if you want to do the right thing, this will require constant vigilance and not necessarily risk free solutions.

Wrapping Up

The AI PM is a really exciting role. You get to be at the field of AI at the best possible moment in history (now!), solve meaningful problems, and deliver real impact. However, it can also be challenging as requires a constant learning curve and dealing with a lot of uncertainty.

As AI continues to grow and shape industries, the need for skilled AI PMs will only increase. If you're passionate about AI and love turning ideas into reality, it's a career worth exploring.