The GenAI Product Manager

The Future of Product Management in the Age of Generative AI.

Author

Thársis Souza

Published

January 23, 2025

Abstract This article examines how Generative AI is transforming the role of Product Management driven by three key trends: the rapidly declining cost of AI capabilities, the increasing anthropomorphization of AI interfaces, and the emergence of human agency. We analyze how these shifts are reshaping traditional Product Management responsibilities, skill requirements, and team dynamics. We also outline common pitfalls to avoid, and explore essential GenAI tools for Product Managers. We argue all PMs will soon be AI PMs but at the same time the core tenets of product management remain more critical than ever, namely, deep problem understanding, business acumen, and user empathy.

Roles

AI PM Roles

Brief high-level definition3 of AI PM roles:

3 Based on Aman Khan’s, Director of Product at Arize AI.

  1. AI Platform PM: Builds foundational AI technology and infrastructure that enables other AI products. Examples include developing APIs for foundation models like Gemini, creating model serving platforms, or building AI development tools.

  2. AI PM: Manages products where AI is the core value proposition to the extent that the product cannot exist without such AI capabilities. Examples include AI-powered image generation, language translation services, or autonomous driving systems.

  3. AI-powered PM: Oversees products where AI enhances and improves existing functionality but is not essential to the core product. The product can function without AI but with AI it functions better/cheaper/faster. Examples include AI-enhanced search recommendations, smart email composition, or automated customer support routing.

I’d argue, all PMs will be AI PMs sooner or later. The more you are focused on AI, the more you will be in the first two roles. The remaining PMs, perhaps the majority, will be in the third role.

The Evolution of Product Team Roles

The impact of AI on product development roles presents two contrasting perspectives leaders are still trying to figure out: One in favor of collapsing role boundaries and another one in favor of maintaining functional excellence.

View 1: Collapsing Role Boundaries

Figure 4: Overlapping roles in the GenAI age

AI is breaking down traditional role silos among product management, design, and engineering - roles that were previously specialized and distinct are now becoming more overlapping towards a more generalist approach - a converging GenAI-powered role (see Figure 4). This perspective suggests:

  • GenAI tools enable individuals to perform tasks previously requiring multiple specialists
  • The traditional handoff model between specialized functions is becoming obsolete
  • Team members increasingly attain cross-functional capabilities
  • “Generalist specialists” who can work across domains are emerging
  • Smaller, more dynamic teams can now handle end-to-end product development
Note

Under this view, in the limit, we would see the first one-person billion dollar company soon emerging4.

4 2025 World Economic Forum, One-Person Enterprise.

View 2: Enhanced Functional Excellence

An alternative perspective emphasizes the risks of role collapse and argues for maintaining functional excellence:

  • While AI enables cross-functional results, faster prototyping and reduces handoffs, core functional expertise remains valuable
  • Product managers risk alienating specialists by attempting to do everything:
    • Engineers still deliver superior engineering outcomes
    • Designers still create superior product experiences
  • Rather than collapsing roles, AI empowers each function to work more effectively, in turn role’s deep expertise makes AI more effective[^deep-expertise].

A Hybrid Approach

I’d argue that the future of cross-functional GenAI-powered roles likely lies between these extremes:

  1. Expanded Capabilities
    • Team members benefit from broader skill sets
    • Understanding adjacent domains improves collaboration
    • All team members become builders
  2. Preserved Expertise
    • Deep functional knowledge remains a value-add
    • Specialists drive excellence in their domains
    • The more expert you are, the more effective you will be in leveraging AI in your domain
  3. Evolved Collaboration
    • Reduced handoffs through prototypes
    • More fluid team interactions
    • The collective impact exceeds individual capabilities

The key is finding the right balance - leveraging AI to reduce friction while preserving the value of specialized expertise. In other words, focus on outcomes over role definitions.

Roles will matter less. We will rather focus on functional excellence that drives business impact, which in turn is fluid and dynamic as the product, user behavior and markets evolve.

Some companies are already experimenting with this approach. For instance, ElevenLabs has no titles and no hierarchy.

Now, there’s no “VP of X”, “Head of Y”, or “Director of Z”. Instead we are just Growth at ElevenLabs, Engineering at ElevenLabs, etc.

Why? We’re small, growing incredibly quickly, and hierarchy just gets in the way.

Instead, the best idea wins no matter where it comes from. And ownership is driven by the results you have.

If you’re obsessed with impact and moving quickly; not titles and status we’re hiring for nearly all roles.

– ElevenLabs employee

AI PM Re-Skilling

Traditional vs. AI Software Products

In order to understand what new skills are needed for AI PMs, first we need to assess how traditional Software Product Development differs from how we will build and manage the next generation of GenAI Software Products:

  • AI vs. User Experience as IP: Traditional software products often relied on proprietary AI capabilities as their key differentiator. With the democratization of AI through foundation models and APIs, unique AI capabilities are becoming less defensible as IP. This shifts the competitive advantage toward superior user experience design - how intuitively and effectively the AI capabilities are integrated into the product workflow becomes the true differentiator.

  • Capability Assessment vs. Functional Testing: Traditional software testing validates specific functionality against predefined requirements. GenAI Product evaluation (Souza 2024), on the other hand, must assess not necessarily pre-defined behavior but also “emergent properties” like reasoning, creativity, and language understanding that extend beyond explicit programming.

  • Metrics and Measurement Challenges: While traditional software metrics can usually be precisely defined and measured, GenAI Product evaluation often involves subjective qualities like “helpfulness” or “naturalness” that resist straightforward quantification. Even when we try to break these down into numeric scores, the underlying judgment often remains inherently human and context-dependent.

  • Dataset Contamination (Xu et al. 2024): Traditional software testing uses carefully crafted test cases with known inputs and expected outputs (e.g., unit tests). In contrast, products leveraging models trained on massive internet-scale datasets risk having already seen and memorized evaluation examples during training, which can lead to artificially inflated performance scores. This requires careful dataset curation to ensure test sets are truly unseen by the model and rigorous cross-validation approaches.

  • Benchmark Evolution: Traditional software maintains relatively stable test suites over time. GenAI models benchmarks continuously evolve as capabilities advance, making longitudinal performance comparisons difficult and potentially obsoleting older evaluation methods.

  • Human Evaluation Requirements: Traditional software testing automates most validation. GenAI product evaluation may demand significant human oversight5 to assess output quality, appropriateness, and potential biases through structured annotation and systematic review processes.

Souza, Tharsis T. P. 2024. “Taming LLMs: A Practical Guide to LLM Pitfalls with Open Source Software.” GitHub Repository. https://github.com/souzatharsis/tamingLLM.
Xu, Cheng, Shuhao Guan, Derek Greene, and M-Tahar Kechadi. 2024. “Benchmark Data Contamination of Large Language Models: A Survey.” https://arxiv.org/abs/2406.04244.

5 In fact, human evaluation/annotation has become a growing business! See scale.ai, snorkel.ai and more.

Aspect Traditional Software Products GenAI Software Products
IP & Differentiation Proprietary AI capabilities as key differentiator Superior user experience design and workflow integration
Capability Assessment Validates specific functionality against requirements May assess emergent properties like reasoning and creativity
Metrics and Measurement Precisely defined and measurable metrics Subjective qualities that resist straightforward quantification
Dataset Contamination Uses carefully crafted test cases Risk of memorized evaluation examples from training
Benchmark Evolution Maintains stable test suites Continuously evolving benchmarks as capabilities advance
Human Evaluation Mostly automated validation May require significant human oversight

Another important implication is that GenAI product evaluation becomes increasingly parametric as each each application may be dramatically different from each other depending on the model, prompt, data, etc as illustrated in Figure 5.

Figure 5: Conceptual overview of Multiple LLM-based applications evaluation.

AI PM Skills

AI Product Management requires a distinct skill set that reflects the unique challenges of developing AI products compared to traditional software. Key competencies include:

User Behavior Analysis

  • Expertise in analyzing and understanding how users adapt their workflows around emerging AI capabilities
  • Skills in measuring and optimizing user adoption, engagement, and retention with AI functionalities
  • Ability to create User Experiences powered by AI, however seamless to AI6

6 “Seamless AI” is a term I use to describe the idea that AI should be so integrated into the product that users don’t even realize they are interacting with AI, i.e. users get the value of AI without the need to think about it.

Capability Assessment and Evaluation

  • Understanding how to evaluate emergent AI capabilities such as reasoning and creativity that go beyond explicit requirements
  • Ability to design meaningful evaluation frameworks for subjective qualities that resist simple quantification
  • Experience with human evaluation processes and annotation workflows

Data and Testing Strategy7

7 I emphasize the importance of Evals. You might ask, where is prompt engineering? Prompt engineering is a new discipline that focuses on how to design effective prompts for AI systems. One could argue it becomes an essential part of the PM job description. Nonetheless, while I agree people often undervalue Prompt Engineering. At the same time, I am not sure whether it will be the PM’s responsibility versus everyone’s job. Going one step further, I’d argue we might delegate prompt engineering to the machines, i.e. we are programming LLMs instead of prompting them. See Stanford’s DSPy.

  • Expertise in dataset curation and validation to prevent contamination issues
  • Understanding of rigorous cross-validation approaches for AI systems
  • Knowledge of evolving benchmarking practices as AI capabilities advance
  • Ability to design comprehensive testing strategies that combine automated and human evaluation

Technical Understanding

  • Sufficient technical depth to assess AI feasibility and limitations
  • Familiarity with AI development lifecycle including data collection, model development, deployment and monitoring
  • Understanding of how AI systems differ from traditional software in terms of determinism and performance characteristics

Process and Risk Management

  • Experience managing highly iterative development processes with frequent course corrections
  • Comfort with ambiguity and ability to make decisions with incomplete information
  • Skills in responsible AI implementation including bias detection and mitigation
  • Ability to rapidly prototype and gather meaningful feedback despite subjective metrics

Continuous Learning

  • Commitment to staying current with rapidly evolving AI capabilities and best practices
  • Understanding of how advances in AI technology translate to product opportunities
  • Knowledge of emerging evaluation methodologies and metrics as the field matures

The role demands a unique blend of technical understanding, process management skills, and strategic thinking to successfully navigate the complexities of AI product development.

Deliverables

Traditional product management has relied heavily on comprehensive documentation like PRDs and test plans to validate functionality and guide implementation. While AI-powered tools now make it faster and easier than ever to create these artifacts, there’s an important caveat: generating documentation should not come at the expense of proper product discovery8.

8 In Appendix A, we provide a brief discussion on how GenAI is changing the nature of product discovery.

Here’s a an overly simplistic view of how the production of some of key deliverables will change, I have seen shared by some PM Leaders:

Before After
Take weeks to write, revise, and sharpen a product strategy 20 minute voice chat with ChatGPT
Take days to consolidate feedback on PRD 15 minutes to scaffold out 80% and then 45 minutes to sharpen
Draw quick wireframes on paper or wait for design to work on UX Share fully functional prototypes with team + customers in minutes
Manually pour through customer feedback to ideas + priorities Automate insight generation with no code tools

While AI tools may accelerate the creation of documentation and deliverables, this view dangerously oversimplifies the complex reality of product management. The most time-consuming and valuable activities - conducting thorough market research, developing deep user understanding, building stakeholder relationships etc - cannot be fully automated and remain critical. These core responsibilities require human judgment, emotional intelligence, and sustained effort to gather genuine insights that drive product success. The time saved on artifact creation should be reinvested in these critical activities rather than viewed as a wholesale reduction in product management effort.

Documentation created without thorough discovery work risks being based on assumptions rather than evidence. The artifacts themselves are not the goal - they should reflect insights gained through rigorous research and validation.

As AI transforms product development, it’s driving several fundamental changes to how we approach these deliverables. I list three key changes likely to happen:

  1. The End of Product Specification Reviews
    • Traditional product spec reviews will end, at least as we know it
    • Teams can quickly create working prototypes to validate ideas iteratively instead of debating specifications
    • Interactive prototypes provide better context and feedback than quasi-static documents
    • AI tools like v0.dev and Cursor allow PMs to prototype features in hours instead of days
    • Real working prototypes lead to more meaningful discussions about product direction
  2. From PRDs to Evaluation Frameworks
    • Traditional PRDs focus on deterministic features
    • AI products require robust evaluation frameworks to assess non-deterministic behaviors
    • Traditional PRDs are requirements-based, AI product requirements are codified by data and evals
    • Evaluation frameworks define success criteria, testing protocols, and quality thresholds
    • The new PRD will be reshaped as a combination of evaluation frameworks and interactive prototypes
  3. The End of User Documentation
    • Arguably, no user reads documentation anymore.
    • Traditional user documentation will be replaced by two key components:
      • For human users: Interactive Product Playgrounds
        • Hands-on environments for users to experiment with features
        • Real-time feedback and learning opportunities
        • Problems/asks/pain points are surfaced by the user through the product itself
      • For AI users: LLMs.txt Specification9
        • Machine-readable documentation for AI systems
        • Standardizes communication between AI and your product
        • Other GenAI tools can automatically integrate with your product via a common LLM friendly specification.
        • This is a reality. See directory of LLM-friendly product documentation here: https://directory.llmstxt.cloud/

9 “The /llms.txt file, helping language models use your website”.

A note on “playground” concept

“Input-optional products”

Don’t ask your users for input. Coming up with input is hard, and a barrier to use. Think of users as wanting to play. We have AI - predict the input! Design products into autonomous environments. Allow users to play by steering a bit.

– Andrej Karpathy

In other words, user problems/asks/pain points should not be assumed to be communicated by the user nor necessarily wait to be discovered by the PM when and if user interviews happen. Instead, the product itself should be designed to be autonomous and able to predict the input!

Having said that, there is a set of product deliverables that will become more relevant as follows:

Product Strategy is King

  • AI capability mapping to business value and user needs
  • User segmentation and value proposition
  • Competitive differentiation strategy
  • Model selection and evaluation criteria
  • Go-to-market strategy for AI features
  • Data strategy and sourcing plans

Evaluation Framework is the New PRD

  • Comprehensive evaluation protocols and acceptance criteria
  • Human evaluation workflows and rubrics
  • Benchmark selection and validation approaches
  • Performance metrics beyond accuracy
  • Quality thresholds for model outputs

User Experience is Everything

  • User interaction flows
  • Prompt design and conversation guidelines
  • Error handling and recovery patterns
  • User feedback collection frameworks

Safety and Alignment is Critical

  • Responsible AI guidelines and principles
  • Content moderation policies
  • Bias detection and mitigation strategies
  • Model behavior boundaries and constraints
  • Safety monitoring and incident response plans

The shift from traditional to GenAI products requires PMs to evolve their deliverables to focus more on evaluation frameworks, user experience design, and responsible AI practices while meeting business needs.

Tools

My personal GenAI PM Stack consists of six key layers that enable Product Managers to effectively build and manage AI-powered products while increasing productivity. Figure 6 provides examples of specific tools for a GenAI PM Stack based on my personal experience - applicability may vary on a per-PM basis.

Figure 6: The GenAI PM Stack. Products included are based solely on my personal experience - applicability may vary on a per-PM basis

Layer 1: Foundation Models The base layer consists of both proprietary and open source models that provide the core intelligence capabilities.

My current favorites:

  1. Proprietary
  • Claude 3.5 Sonnet: Best at writing
  • GPT-4o-mini: Good reasoning / per unit of cost ratio
  • Gemini: Overall cheapest/fastest high-end model also best at multi-modal and long-context tasks
  1. Open Source
  • DeepSeek: SOTA Open Source LLM
  • Qwen: Best Open Source LLM per unit of size
  • Llama: Rich open source ecosystem / Best for Enterprise customization

Layer 2: Model Inference & Serving This layer provides the infrastructure to actually run and interact with foundation models, locally. It includes both command-line tools for developers and user-friendly interfaces for non-technical users.

My current favorites:

  1. Inference
  • LLama.cpp: Highest performance for local inference
  • Ollama: Best user-friendly way to experiment locally with Open Source models
  • LLamafile: Best for maximum portability
  1. UI
  • Jan: Simple and accessible alternative to ChatGPT aimed at non-technical users
  • LM Studio: Easy multi-model local serving (closed-source)
  • OpenWebUI: Open Source alternative to LM Studio also a fit for enterprise usage

Layer 3: LLM Tools & Frameworks The tools layer offers frameworks and testing capabilities essential for building production LLM applications. This includes options for orchestrating LLM calls, managing complex workflows, and evaluating model performance.

My current favorites:

  1. Building LLM applications:
  • LangChain: Most popular framework for building LLM applications
  • LangGraph: Best for building LLM applications with complex workflows including agents
  • LLamaIndex: Best for building LLM applications with complex data including RAGs
  1. Testing LLM applications:
  • PromptFoo: Best for testing LLM applications with complex prompts
  • LightEval: Seamless integration with Hugging Face models, datasets and metrics
  • LangSmith: Best for observability and debugging LLM applications, also for Evals

Layer 4: Domain-Specific Tools This layer contains specialized tools for specific AI capabilities like text-to-speech, avatar generation, and image/video creation - enabling multimodal AI experiences. This layer is highly dependent on the product and the domain.

My current favorites:

  • ElevenLabs: Text to speech
  • HeyGen: Avatar generation
  • MidJourney: Image Generation
  • Sora: Video Generation

Layer 5: IDEs & Development Tools This layer focuses on tools that enhance the development experience through AI assistance, from code completion to full application generation10.

10 The Pragmatic Engineer newsletter has a great list of IDEs with GenAI features.

My current favorites:

  1. IDEs & Code Editors
  • Cursor11: Best overall AI-powered IDE
  • WindSurf: Main competitor to Cursor
  • GitHub Copilot: The first and most used AI coding assistant

11 Cursor has just raised $105m in Series B funding.

  1. Low/No-Code Tools
  • v0.dev: Best for UI prototyping
  • bolt.new12: Best for full-stack application generation

12 bolt.new announced days ago a $105.5m funding round.

Layer 6: Productivity Tools The top layer consists of AI-powered applications that enhance day-to-day productivity, from writing assistants to research tools, making AI capabilities directly accessible to end users.

My current favorites:

  • Claude 3.5 Sonnet: Best for writing
  • o1: Best for planning
  • Gemini: Best for speed, cost and multi-modal
  • NotebookLM: Best for user-provided grounded research
  • Gemini Deep Research: Best for web-based grounded research
  • Notion: Best for knowledge management
  • Granola: Best for AI-powered note taking
  • Perplexity: Search

Pitfalls

As GenAI technologies advance coupled with the pervasive belief that AI will solve all challenges, there is a risk of over-reliance on AI systems. This risk is exacerbated as boundaries of what’s human versus AI responsibility becomes increasingly unclear, leading to oftentimes excessive delegation of critical thinking and decision-making to machines. Key pitfalls include but are not limited to the following:

  1. Over-relying on AI-Generated Content
    • One of the most seductive traps in AI Product Management is excessive reliance on AI tools for generating product artifacts without sufficient human validation and discovery work. While AI can accelerate documentation creation, using it as a substitute for genuine user research and thoughtful product strategy leads to superficial solutions that fail to address core user needs.
  2. Product Demo is not Product Production
    • Product demos are great for showcasing the product, minimizing handoffs and collecting feedback early-on. However, they are not a substitute for production code. In fact, demos can be a distraction from the actual product development if production requirements are overlooked, such as observability, and reliability.
  3. Alienating Cross-Functional Teams
    • As PMs become more technically proficient with AI tools, there’s a risk of overstepping boundaries and alienating engineering teams by attempting to generate code or make technical decisions without proper collaboration.
    • Instead, PMs should leverage AI tools to enhance cross-functional collaboration - using them to better communicate requirements, create clearer specifications, and facilitate more productive technical discussions.
    • The goal is to use AI as a bridge for better collaboration, not as a replacement for essential cross-functional partnerships and expertise.
  4. Misunderstanding AI Capabilities
    • There is often an optimism bias in the technology community that AI can solve everything (better/faster/cheaper). The other side of the coin is also true, i.e. lack of understanding of AI stack can lead to under-utilization and missed opportunities.
    • More and more technical expertise will become part of the PM job description to avoid “Black Box Thinking”, i.e. treating AI as magical rather than understanding its limitations and constraints as well as potentials to close the gap between AI promise and reality.
  5. Late Evals
    • Another common pitfall is late evaluation planning, which occurs when the design of the evaluation framework is postponed until after implementation. This delay makes it challenging to measure effectiveness and can result in missed expectations.
    • To address this, the evaluation framework should be designed early in the process and integrated throughout the development cycle.
  6. Weak Evals
    • It is common to begin with simple evaluations that focus on a single dimension of requirements, and that’s a good approach: start simple, iterate, learn, and improve. However, the real mistake occurs when these initial checks are not evolved throughout the development cycle.
    • As a consequence, teams might have a sense that performance results are strong when in reality it might be data evals are weak, instead.
    • Before moving to production, it is crucial to establish well-balanced datasets that represent requirements in a nuanced manner better reflecting real-world user scenarios.

Takeaways

  • All PMs will be AI PMs.

  • Despite the allure of AI tools and technologies, nothing beats being passionate about the problem and maintaining an obsession with understanding users.

  • AI is changing the way we create products, the key is to understand user behavior change. Product discovery and evals will be critical.

  • AI won’t replace humans, but humans augmented by AI will replace humans who ignore AI’s evolution.

Open Questions

  • How might AI change how value is measured, attributed, and captured – not to mention created?
  • At what point should AI PMs design products with LLMs as the intended users?
  • How will the role change if LLMs formally take on some of the stakeholder roles (e.g. customer success)?
  • What sort of advance, if it happened, would make this article’s assumptions wrong or irrelevant?

Acknowledgements

I’d like to thank Pam, Sam and Jordan for their feedback and suggestions on this article.


Appendix A: How Product Discovery is Changing

How will generative AI change the nature of product discovery? We don’t know yet. But there is a recent study by MIT researcher Aidan Toner-Rodgers (Toner-Rodgers 2024) that might provide insights into how GenAI will affect discovery for knowledge works.

Toner-Rodgers, Aidan. 2024. “Artificial Intelligence, Scientific Discovery, and Product Innovation.” https://aidantr.github.io/files/AI_innovation.pdf.

13 The AI tool works by generating “recipes” for novel compounds based on specified properties, i.e. “Inverse Material Design”, the researchers input desired proprieties and through Graph Neural Networks the AI tool generates candidate materials. Scientists then evaluate and synthesize the most promising candidates. The AI automates a significant portion of the “idea-generation” tasks (57%), reallocating researchers to evaluating model-produced candidate materials

This work examines the effects of a new AI-powered materials discovery tool on the productivity, creativity, and job satisfaction of 1,018 scientists13. Findings are striking:

  • AI-Driven Productivity Gains: The introduction of the AI tool led to a significant increase in research output. AI-assisted scientists discovered 44% more materials, resulting in a 39% increase in patent filings, and a 17% rise in downstream product innovation. These compounds also possessed more novel chemical structures and led to more radical inventions

  • The Importance of Human Judgement: The paper investigates the reasons for the disparate impact of AI, highlighting the importance of human judgement. The study shows that scientists with strong judgement skills are better able to prioritise promising AI suggestions, whereas others waste significant resources testing false positives. This indicates that AI does not diminish the need for human expertise, but rather changes the type of expertise that is valuable

  • Impact on Job Satisfaction: Despite the productivity gains, the study found that 82% of scientists reported reduced satisfaction with their work due to decreased creativity and skill under-utilization. This suggests that AI implementation must consider the impact on job satisfaction, and that workers may require additional training and support as they adjust to the changing nature of their work

The study provides a good starting point for thinking about the implications of AI in discovery, in general, and product discovery, in particular. It’s important to keep in mind that Product Management is a discipline with its own unique set of challenges. However, I’d argue some key points should be considered as we augment the product discovery process with AI:

  • Human-AI Partnership:
    • Complementary Strengths: While AI can automate ideation and concept generation, human product managers remain essential for evaluation, strategic direction, and contextual understanding. PMs become orchestrators of AI capabilities rather than being replaced by them.
    • Value-Add Focus: PMs should focus on areas where human judgment adds the most value - strategic decisions, user empathy, and complex trade-offs.
  • Shift in Role Focus: Product managers will likely shift from traditional ideation and brainstorming towards evaluating and prioritizing AI-generated product concepts, features, and designs. The emphasis moves from creation to curation.
    • Decreasing in Importance: Pure ideation and brainstorming skills may become less critical as AI can help automate idea generation.
    • Growing in Importance: Judgment and evaluation capabilities become paramount. PMs need strong abilities to assess ideas and scenarios with deep domain knowledge to evaluate their viability. Understanding and interpreting AI outputs becomes a core competency.
  • Enhanced User Research: As AI can rapidly generate multiple product concepts and solutions, the importance of rigorous user research grows significantly.
    • Validation Critical: With AI generating numerous possibilities, PMs must excel at validating which solutions truly resonate with users through careful research and testing.
    • Deep User Understanding: The ability to deeply understand user needs, pain points, and contexts becomes even more crucial to guide AI-assisted product development in the right direction.
    • Research-Driven Iteration: User research becomes the key driver for iterating on AI-generated solutions and ensuring they solve real user problems effectively.

The materials science study raises important questions about how AI-augmented product discovery affects PM job satisfaction and creativity. Some important questions remain open:

  • Redefining Creativity: How can we meaningfully distinguish between human PM creativity and AI-assisted product discovery? What unique creative value do PMs bring that AI cannot replicate?
  • Job Satisfaction Impact: As AI increasingly automates ideation and concept generation, how will this affect PM job satisfaction and sense of creative ownership? What new sources of fulfillment might emerge in an AI-augmented discovery process?
  • Balancing Automation and Agency: What is the right balance between human and AI capabilities?

Appendix B: Resources

There is an overwhelming amount of fragmented information channels and resources at the intersection of GenAI, Business, Product, and Technology. From new online PM courses (released almost daily) on Maven14, to podcasts15, VC-curated resources16, to PM-specific AI tools17.

14 Product Sense by Shreyas Doshi is perhaps the most popular online general PM course but new GenAI specific courses are released almost daily such as Build With LLMs or How to Get Into AI.

15 Examples of good Podcasts in the area include No Priors with VCs Elad Gil and Sarah Guo, and Generative AI Now. But the most popular one by far is Lenny’s.

16 a16z often releases relevant AI resources at https://a16z.com/ai/.

17 Examples include LennyBot and ChatPRD.

However, rather than providing an exhaustive list of conventional resources, this section focuses on three enduring pathways for developing expertise in AI Product Management, namely learning through Building, Community, and Foundations.

Note

The best resource for aspiring AI Product Managers is the combination of hands-on building experience, active community participation, and solid foundational knowledge. This trinity provides the practical skills, network, and strategic understanding needed to excel in this rapidly evolving field.

Learning Through Building

The most effective way to develop AI PM skills is through hands-on experience building AI-enabled products. I can’t emphasize this enough. Finding a problem you’re passionate about and using it as a learning vehicle is the best way to develop both technical intuition and product sense simultaneously.

Developing for yourself, as the first user, almost guarantees you are motivated to solve the problem and is a great way to increase the likelihood you deeply understand the pain points and frustrations associated with that problem.

Once you become obsessed with the problem, you start walking backwards from the problem to the solution and that’s when you start to understand the technical constraints and possibilities. Learning GenAI capabilities18 comes as a consequence of the problem solving process.

18 Generative AI for Beginners by Microsoft is a great resource for getting started with building with GenAI.

Another consequence is that you start to learn from others who are solving the same or adjacent problems, which brings the second point: Community Engagement.

Community Engagement

Here, there is no better resource than the open-source AI community. In my experience, that’s a community that is willing to help and share knowledge while being very open to new ideas and approaches. You will find:

  • Active Learning Communities: Platforms like Hugging Face host vibrant communities where product managers can engage with developers, researchers, and fellow PMs.
  • Real-time Knowledge Exchange: Open source projects provide visibility into cutting-edge developments and implementation approaches. You can simply make a contribution to a project and get feedback from the community.
  • Network Building: Engaging with the community creates connections with others working on similar problems and technologies. This is also applicable at the enterprise level where private companies continue to value open-source contributions.

Foundations

While Generative AI offers higher levels of abstraction for problem-solving, understanding fundamental concepts remains crucial for strategic product management:

  • Technical Understanding: Knowledge of AI/ML foundations helps understand today’s and tomorrow’s capabilities and limitations which in turn helps PMs make better strategic decisions.
  • Strategic Foresight: Foundation knowledge improves ability to identify which problems will remain relevant given AI’s rapid pace of development. PMs should invest in problems that will become more, rather than less, relevant as AI capabilities advance.
  • Risk Assessment: Technical foundations enable better evaluation of AI-related risks and limitations. These include Safety, Bias, and Alignment requirements but also technical constraints such as latency, cost, and scalability.

I’d recommend the following resources to get started:

Appendix C: DeepSeek-R1

Here’s DeepSeek-R1’s answer to the question: “Will Product Managers survive the Generative AI revolution?”.

The role of Product Managers (PMs) will evolve rather than disappear in the face of the Generative AI revolution. Here’s a structured analysis of how PMs will adapt and thrive:

Augmentation, Not Replacement

  • Data-Driven Insights: Generative AI can process vast datasets to identify customer needs, market trends, and feature priorities, but PMs will interpret these insights within strategic and ethical contexts.
  • Efficiency Gains: Automating administrative tasks (e.g., report generation, documentation) allows PMs to focus on high-value activities like vision-setting and stakeholder collaboration.

Human-Centric Skills Remain Critical

  • Strategic Vision: Aligning product goals with business objectives requires creativity and intuition—areas where human judgment outperforms AI.
  • Cross-Functional Leadership: Building trust, negotiating priorities, and motivating teams rely on emotional intelligence, which AI cannot replicate.
  • Ethical and Empathetic Decision-Making: Navigating complex moral dilemmas and ensuring inclusivity demands human oversight.

Evolution of the PM Role

  • AI Collaboration: PMs will increasingly act as “translators” between AI outputs and actionable strategies, leveraging tools to enhance decision-making.
  • New Opportunities: Managing AI-driven products (e.g., generative models, ethical AI systems) will require PMs with technical literacy and market expertise.
  • Focus on Innovation: Freed from routine tasks, PMs can prioritize experimentation, long-term roadmaps, and disruptive ideas.

Industry Parallels

  • Job Transformation: Like ATMs for bank tellers, AI will shift PM roles toward strategic advisory and creative problem-solving rather than eliminating them.
  • Demand in AI-Centric Sectors: As AI products proliferate, PMs with expertise in AI governance, user experience, and scalability will be in higher demand.

Conclusion Product Managers who embrace AI as a collaborative tool—while honing uniquely human skills like leadership, empathy, and strategic thinking—will not only survive but lead innovation. Adaptation is key: PMs must upskill in AI literacy, ethical oversight, and cross-functional collaboration to thrive in an AI-augmented future. Those resistant to change may struggle, but the role itself is poised to grow in significance as products become more complex and AI-dependent.

Citation

BibTeX citation:
@online{souza2025,
  author = {Souza, Thársis},
  title = {The {GenAI} {Product} {Manager}},
  date = {2025-01-23},
  url = {https://www.souzatharsis.com/writing/aipm},
  langid = {en}
}
For attribution, please cite this work as:
Souza, Thársis. 2025. “The GenAI Product Manager.” January 23, 2025. https://www.souzatharsis.com/writing/aipm.