Proof over Persuasion

My first internship was at a brand management company, the original school of product craft.

Before tech or growth hacking, brand managers owned products end to end. Research, design, pricing, distribution, advertising — all of it. That was the first time I saw what real ownership looked like: living inside the product, knowing both the numbers and the narrative.

Those early lessons taught me something that’s still true today: good products are built on proof, not persuasion. The best brand managers had both rigor and taste. Data showed what had worked; craft shaped what would. That blend of evidence and intuition became the foundation for how I think about product.

When I moved into technology, the idea of ownership stayed but the medium changed.

At Facebook, reviews were moments of truth. You showed something real, not slides. If it clicked, the room could feel it; if it didn’t, you learned quickly. Boz’s P Rules were part of that culture and a shorthand for how to keep reviews honest. No piling on (don’t repeat someone else’s critique just to agree), no pedantry (don’t nitpick to look smart), no prescription (don’t dictate fixes), and no permission (don’t wait to be told what’s right). The rules reminded us that the work should speak for itself, and that leadership’s job is to see clearly, not to perform expertise.

When we built fundraising products, those values were tested. We were designing for generosity, not attention. Every choice had to be simple, sincere, and grounded in empathy. The only real measure was whether someone decided to donate; proof you couldn’t fake or misinterpret.

On Messenger and later in video, the challenge to solve for was presence. The product was about connection, tone, and trust; things that couldn’t be explained in words. A few milliseconds of lag could break a conversation. No amount of storytelling could fix that. Proof lived in the experience.

At Instacart, the pressure shifted from speed to reliability. Grocery delivery became essential. Product reviews turned into command rooms where every bug mattered. The metric that mattered most was recovery time. Function was the design, and proof was performance under pressure.

All these experiences reflect the same lesson in different dialects. Empathy, connection, reliability; each changed what proof looked like, but not what it meant. The best reviews stayed grounded in evidence and honesty.

Once you experience reviews that run on proof, the anti-patterns are easy to spot: rooms that reward polish over progress, leaders who seek certainty instead of signal, teams who prepare arguments instead of prototypes. It is easy for a review to become theater. It is much harder to keep it a workshop.

My goal is not to eliminate structure, but to make structure serve truth. The best review systems surface data, show the product live, and invite genuine feedback. They make the product the centrepiece, and balance logic and taste, measurement and meaning. Craft is care, and proof that someone took the time to understand what they are building and why.

That principle matters even more now as we build products shaped by AI. These systems learn, adapt, and surprise. A review cannot be a snapshot in time, it has to be a reflection loop. We open the product, run real prompts, look at outputs, and ask what the system learned. Proof lives in telemetry and traces, not in plans or slides. We evaluate precision and judgment, but also tone and coherence. It’s about how the system performs and how it feels.

Within Microsoft CoreAI, we’re putting this to practice. Our reviews bring live experiments, system metrics, user signals, and direct demos. We bring failures and edge cases, not just successes. In one round, we learned that small latency improvements boosted user satisfaction more than accuracy gains. It’s a reminder that we need to study both the quantitative and the qualitative, or else we miss part of the truth. The data tells us where we are; the design shows us where to go.

Younger or smaller companies often start this way naturally. For those at scale, the shift is harder. It means unlearning habits built around persuasion. Choosing to walk into rooms without polished answers and invite others to see the product as it is. This kind of honesty is uncomfortable, but it’s the only way to see what’s real. Reviews should be mirrors, not approvals.

I see craft and data as partners. Data keeps us honest. Craft keeps us human. Design turns both into something people can feel.

We will keep showing our work and learning from it; proof over persuasion. That’s how you build products that last, and cultures that tell the truth.