I once launched a product that I thought would change the world. It ended up burning me out and, worse, it failed. Not because the tech wasn't good enough, but because I was fundamentally wrong about what it means to build responsibly with AI.
We had built a sophisticated recommendation engine, let's call it "ConnectSphere." The idea was to connect people in large communities based on deep, shared interests, not just superficial likes. The algorithm was brilliant. On paper, it was a masterpiece of collaborative filtering and NLP. Our engineers were some of the best I'd ever worked with. We spent months fine-tuning models, debating fairness metrics, and running bias tests. We thought we had “ethical AI” covered.
But we were asking the wrong questions. We were so focused on the technical perfection of the how that we completely missed the human impact of the what. The product worked exactly as designed. It created tight-knit groups. So tight, in fact, that they became echo chambers. It discouraged new ideas and diverse perspectives. We didn't build a community builder; we built a digital fortress. Engagement looked great for a while, but the user experience was toxic. The burnout came from trying to fix a human problem with engineering solutions after the fact. It was a painful, expensive lesson.
That failure forced me to see the truth: ethical AI is a product management problem.
The Illusion of the Engineering Fix
For too long, we've treated AI ethics as a technical challenge. A bug to be patched. We talk about debiasing datasets, ensuring model fairness, and creating explainable AI (XAI). These are important, but they are table stakes. They are the engineering team's responsibility to implement a solution correctly.
They are not, however, a substitute for a product manager's responsibility to define the right problem to solve in the first place. An algorithm can be technically “unbiased” and still be used in a product that causes immense harm. A perfectly fair model that allocates predatory loans is still a predatory product. The code is just executing instructions. The product manager writes the instructions.
As a PM, you are the conscience of the product. You are the first line of defense. Delegating that responsibility to a checklist or an engineering team is an abdication of your core duty.
The Shift: From Model to Mission
The perspective shift that changed everything for me was this: stop focusing on the model and start focusing on the mission. What is the product’s purpose in a human being's life? What impact are you trying to create? Who benefits and who might be harmed?
This is product management 101, yet it gets lost in the hype of AI. As an investor, I see hundreds of pitches a year. The ones that stand out aren't just about a technical edge. The companies I've been fortunate enough to back, like Anthropic and Scale AI, are deeply thoughtful about the purpose of their technology. They have product leaders who are obsessed with the potential for both good and bad outcomes. They ask the hard questions from day one, not as an afterthought when something goes wrong.
Your job as a PM in the AI space is to be the chief skeptic and the chief humanist. It involves:
Defining “Good”: “Ethical” is a vague term. For your product, what does it mean? Does it mean user well-being? Fair allocation of resources? Creative expression? You must define it in a way that can be debated, measured, and designed for. It's not a universal constant; it's a product-specific choice.
Pre-Mortems for Harm: We all do pre-mortems for why a product might fail to hit its business goals. We need to be just as rigorous in running pre-mortems for how a product might cause harm. What are the worst-case scenarios? How could this be misused by bad actors? What are the unintended consequences for individuals and society? Brainstorm these with the same intensity you brainstorm new features.
Expanding Your Metrics: If your only success metric is daily active users or time-on-site, you are optimizing for addiction, not value. What are the metrics for a life well-lived? For a user feeling more connected, more knowledgeable, or more secure? These are harder to measure, but they are not impossible. Finding proxies for human well-being is the next frontier of product management.
This Is Your Job Now
Building with AI is not the same as building a simple SaaS app. The scale, speed, and autonomy of these systems create new and amplified risks. You cannot simply A/B test your way to a responsible product.
It requires a deep sense of ownership. It means having uncomfortable conversations with leadership about prioritizing safety over short-term growth. It means building diverse teams who can spot blind spots you can't see. It means being brave enough to say “no” to a feature or a product that you believe will cause more harm than good.
Stop waiting for the engineers to solve the ethics problem. They can’t. They are waiting for you to define the product, to set the boundaries, to articulate the mission.
Ethical AI is not a separate discipline. It is product management. Own it. The success of your product—and your conscience—depends on it.
Frequently Asked Questions
Do all experts agree with this view?
No, and that's fine. The best ideas in business are often contrarian. I share my perspective based on my experience and data, but I encourage you to seek out opposing viewpoints and form your own conclusions.
What experience informs this perspective?
This perspective comes from over a decade of building companies in Silicon Valley, two successful exits (RemoteTeam to Gusto, MovieLaLa to Gfycat), and investing in 200+ startups including Anthropic, OpenAI, and Scale AI. I write about what I've lived.
What's the most common pushback you get on this?
People often push back by citing exceptions or edge cases. And they're usually right that exceptions exist. But building a strategy around exceptions rather than patterns is a losing game for most founders.