Before GenAI, it was already hard to explain what exactly product managers do.

We’re not engineers, but we work on the product.

We’re not designers, but we help define the user experience.

We’re not in sales, marketing, or customer success, but we think constantly about go-to-market and how it lands with customers.

Over the years, I’ve found the simplest way to explain product management, especially to people outside of tech (like my parents), is this:

We define what needs to be built, why it matters, and how it drives the business forward.

That “what” and “why” still hold in the GenAI era.

But the “how”? That part is definitely changing.

The “how” is getting rewritten

AI-powered features don’t behave like traditional software. They’re not deterministic. They generate outputs that are fluid, context-sensitive, and often unpredictable.

This means the “how” of product management is shifting. It’s no longer just about defining logic flows or UI behaviors, it’s about shaping system behavior across a wide range of user intents, inputs, and scenarios.

• Behavior is influenced by prompts, tools, and context, not just code.
• User inputs are open-ended, which makes it impossible to predefine every outcome.
• Success depends on how well the system adapts to real-world usage, not just whether it matches a spec.

The challenge? Our old definition of “done” doesn’t quite hold up. Instead of verifying that a feature behaves exactly as written, we’re now evaluating response quality, behavioral patterns, and whether the AI is meeting user goals across varied contexts.

That’s why specs can’t be static anymore. They need to evolve as the product does. PMs must stay engaged throughout the entire lifecycle: shaping prompts, refining context, defining evaluation criteria, and adjusting based on actual usage.

AI success depends on understanding the users, not just the model

Let’s be honest, even in traditional software, defining “success” wasn’t always clear-cut. With GenAI, the ambiguity is just more visible.

We’ve always needed PMs to define what needs to be built and how. But now that AI can seemingly generate anything on its own, it’s easy to assume we no longer need deep user or market understanding. That’s a dangerous misconception. In reality, we need it more than ever because AI’s output can look polished and plausible while still being completely wrong or irrelevant.

Consider an AI-powered budgeting tool. The system might involve agents, tools, even multi-step orchestration through a financial platform. But if you don’t understand how people actually budget, how they categorize expenses, plan for goals, or react to certain tradeoffs, you won’t know when the AI is giving bad advice. Worse, it may sound convincing while quietly misfiring. In a world of probabilistic systems, knowing when it’s not working is harder than ever and that’s exactly where PMs come in.

While it might look like AI can generate everything on its own, PMs still need to represent the user throughout the development process. AI should be guided by problem solvers, not just builders. Without intentional direction, systems risk optimizing for what’s technically possible instead of what’s valuable, usable, or aligned with real human needs.

That doesn’t mean PMs can ignore the tech. We still need enough technical fluency to collaborate effectively on model behavior, tools, and system design. But the core superpower remains the same: bridging human context with product capability.

Bottom line: even if it feels like you can “just tell AI to build something,” real product success still depends on clear framing, structured iteration, and sharp product judgment.

Looking ahead: Adapting to the new product reality

The rise of GenAI isn’t just changing what we build, it’s reshaping how we build it. As product managers take on more responsibility for guiding AI behavior, we need to evolve our processes:

• Iterate through uncertainty: We’re not refining deterministic features – we’re shaping behavior in a probabilistic system.
• Redefine specifications: Specs now serve as living frameworks: anchoring intentions, outcomes, and guardrails instead of strict logic.
• Reimagine quality: Success is about behavior alignment, not binary outputs. PMs need to help define what good looks like across messy, real-world inputs.
• Evolve monitoring: AI systems need new feedback loops: how users engage, how outputs vary, and when behavior drifts.

The strongest teams will build new collaboration patterns between PMs and engineers, ones that embrace exploration, iterative alignment, and continuous tuning.

So, where do you stand in this shift?

Are your product processes still built for a deterministic world?

Or are you evolving with the new reality of GenAI, where AI’s value depends not just on what it can do, but what we help it do well?

While AI may change the how, it doesn’t replace the why or the people who understand it best.