AMA: Capital One Director, Product, Mona Salvi on AI Product Management

Hiring managers are increasingly thinking about AI in two complementary ways when evaluating candidates: AI‑enhanced skill sets: Candidates who can leverage AI to amplify their work whether that’s data analysis, content generation, process automation, or product insights. Understanding how to use AI responsibly, strategically, and efficiently is becoming a baseline expectation in many roles. Human strengths AI won’t replace: At the same time, managers value uniquely human capabilities: critical ...Read More

The conversation around AI is increasingly converging on a shared-accountability model. AI agents cannot be legally liable on their own, accountability always traces back to the humans and organizations that design, deploy, and authorize them. As autonomy increases, expectations rise for responsible governance, strong guardrails, and continuous oversight. If an AI agent makes a bad decision, the liability doesn’t belong to the model, it belongs to the people who gave it authority. In practice, t ...Read More

As a CyberSecurity Product leader, I am closely monitoring emerging AI and machine learning trends that could directly enhance security, efficiency, and product trust. A few areas that I am closely watching are: Adversarial AI: Understanding how models can be intentionally manipulated helps us anticipate vulnerabilities and build more resilient AI-driven products. Agentic AI for pentesting: Autonomous AI agents are beginning to simulate attacks, identify weaknesses, and test defenses at scale, e ...Read More

Managing safety, security, and compliance for autonomous AI isn’t fundamentally different from how we approach secure coding or product security today, it’s the same mindset applied to systems that can act autonomously. A source I strongly suggest organizations to reference as they design AI systems is the NIST AI Risk Management Framework (AI RMF 1.0). It provides a practical structure for this approach, emphasizing that trustworthy AI should be safe, secure, and resilient and accountable and t ...Read More

It really depends on the problem you are trying to solve. Some examples below. However across all domains, traction generally comes from: Augmenting human expertise, Automating repetitive work, and Delivering actionable, personalized insights to users so that users can time back to do more important things in their day. CRM / Marketing: The biggest traction is in personalized content generation (emails, social posts, landing pages), sales enablement (drafting follow-ups, summaries, proposals), a ...Read More

Shipping agents will be table stakes. However winning will be for those who build keeping trust, context awareness, and accountability in mind, not just autonomy alone. For example: tool and system integration, autonomous task execution, remembering user preferences, personalizing context and content - these are tablestakes, great features, required features! True differentiation however will come from:Trust and reliability (Before executing a risky action, the agent flags uncertainty and reques ...Read More

Certifications and training can help, however certifications alone are not enough. In my opinion, staying ahead means applying AI strategically, understanding risks, and iterating quickly, just like any other emerging marketing technology. Being a security practicioner myself, here are some security specific AI courses/certifications that I have advised students in my other seminars (Eg: San Jose State University). Thesev courses will help understand: AI‑specific threat models and risk vectors G ...Read More

To be effective in AI strategy and product management, PMs need to build skills that go beyond just managing work for engineers. Great PMs need to exhibit skills that showcase strategic thinking, risk evaluation and ability to translate AI potential into product value. Key areas include: AI Literacy: Understanding core AI concepts, capabilities, and limitations, enough to evaluate feasibility, ask the right questions, brainstorm with Engg and architects, work with first line risk, executive lead ...Read More

My #1 mantra for successful products: Know your customer, focus on your customer pain points and leverage data from the pain points to priroitize things that matter most! When it comes to balancing model metrics and user feedback: metrics tell you if the model works, but user feedback tells you if it truly matters. When they conflict, I always try and dig into the root cause, is it a data gap, a UX issue, or a misalignment with the product goal? Iteration, experimentation, and careful analysis h ...Read More

As AI increases code output, our security programs need to go beyond tracking coverage and count of vuklknerabilities into understanding real exploitability and impact. Traditional coverage metrics tell us what has been exercised, but they don’t tell us what attackers can actually exploit or what would meaningfully harm customers or the business. In the AI era, we need to shift toward risk‑based security signals that focus on the threats that matter most. We are already seeing examples of this i ...Read More