At Operator Collective's Spring Gathering, a roundtable on AI board governance drew together current and former board members, executive search and board placement leaders from Egon Zehnder, and senior operators to work through one of the most consequential questions in enterprise leadership right now: how should boards govern AI without slowing down the companies they're meant to guide?
The board AI talent question is being asked too narrowly
Many companies are approaching board AI expertise the same way they approached cybersecurity expertise a decade ago - find one person who knows the subject, put them on a committee, and assume the topic is covered. That approach alone isn’t enough for AI. Unlike many previous technology shifts, AI has implications for every aspect of the enterprise from strategy, product development and customer experience to operations, talent and risk.
We see three distinct profiles that boards are looking for right now, each with a different purpose:
- The first is an AI-proficient, product and technology-first leader who can advocate credibly for how AI and technology can be deployed to drive huge value for customers, build competitive advantage and help management teams think more boldly about what is possible.
- The second is a genuine AI expert with deep technical expertise, which is the right fit for some companies but is not necessary for all.
- The third, and perhaps most underappreciated, is an experienced, bold decision-maker who has successfully led through scale and change. Someone who has navigated the kind of high-stakes enterprise decisions that management teams are working through right now, and can offer not just technical knowledge but can help with strong judgment and principles at scale under uncertainty.
This is about understanding what is possible. Many boards are embedding AI discussions across the board and committees rather than treating AI as a standalone topic owned by one director or one committee, just like the strongest companies have both central AI resources and AI proficient leaders embedded across the company. More boards are also dedicating time and budget to AI education for existing directors, recognizing that fluency can be built and that the cost of not knowing is higher than the cost of training.
The questions boards should be asking but often aren't
The top questions that boards should be asking are split into two categories:
The first category is existential.
- If a well-funded startup entered your market today and built from scratch, what would they do differently and what competitive advantage would that give them?
- If you were starting your company today with full knowledge of what AI can do, what would you do differently? What problems would you solve for customers that you're not solving today?
- What if, in 12–18 months, more than half of our customers were interacting with us through their own AI agents rather than directly? What should we be doing now to prepare for and win in that world?
These are not predictions. They are planning questions designed to help management teams and boards think proactively about a range of possible futures.
The second category is operational.
- How does the company move from being a system of record to being an intelligent system of action?
- What is the build, buy, and partner framework for AI capabilities, and is the board regularly reviewing principles and progress here?
- How is progress on AI being measured, from initial metrics like daily active use, PR cycle time and token use to more advanced, value-driven metrics like productivity gains and business outcomes?
- As AI systems become more capable on both sides of the security equation, is the company's cybersecurity posture keeping pace with the sophistication of the attacks it is facing or will face?
AI has become a standing board topic at many companies. The challenge now is ensuring those discussions move beyond awareness and updates on experiments to important strategic decisions and measurable outcomes.
What good board AI governance looks like
On reporting: AI goals, key metrics, and risks should be reviewed at every board meeting, not every 6 or 12 months. That means usage data, outcomes, challenges, wins, and what the organization has learned. The build, buy, and partner framework for AI should be established and reviewed regularly for progress. Benchmarking against peers inside and outside the industry should be part of this, because AI maturity is relative and boards need context to evaluate what they're hearing.
On education: we are emphatic that hands-on time dramatically outperforms presentation-based learning. The companies and boards that have seen the most meaningful shift in AI fluency are the ones that have moved from slide decks to actual use, working hands-on with tools, building prototypes and running agents. Weekly sessions where employees show their teammates what they've actually built and how it's helping them are consistently more effective at driving adoption than watching general purpose presentations. The understanding you get from using something yourself is just different from the understanding that comes from being told about it.
The governance question that deserves more attention
One question kept coming up. How can boards govern AI in a way that helps companies move quickly and manage risk? Companies face two risks simultaneously - moving too slowly and missing important opportunities or moving too quickly without appropriate controls and quality.
The companies making the most progress have moved beyond viewing AI solely through a risk management lens. Leading boards are approaching AI both as a governance topic and a strategic capability. The boards making the most progress are building the capability to ask good questions, support thoughtful decision-making and help management teams prepare for opportunities and challenges that are arriving more quickly than ever before.
This article was adapted from a roundtable discussion at Operator Collective's Spring Gathering, hosted by Egon Zehnder and held under Chatham House rules. Insights are shared without individual attribution.
