What Keeps Me Up at Night as a Director of an AI Company

I serve on the boards of companies that build AI. I also serve on boards of companies that use AI. The two experiences are very different, but they share one thing in common: the questions keep getting harder.

Let me share what I have learned — and what keeps me up at night.

The Speed Problem

The first thing that strikes you as a director in an AI company is the pace. Traditional board cycles assume quarterly reporting, annual strategy reviews, and multi-year planning. AI development cycles operate in weeks and months.

A product that was compliant and safe at the last board meeting may have changed fundamentally by the next one. The model may have been retrained on new data. The deployment scope may have expanded. The regulatory environment may have shifted.

This creates a tension that boards are not designed to manage: how do you exercise oversight over something that changes faster than your meeting schedule?

What I have learned: You need a different cadence. Not more board meetings — but better information flows. Boards need real-time or near-real-time visibility into material changes, not just quarterly reports. This requires a different relationship with management and a different approach to reporting.

The Explainability Problem

The second thing that keeps me up at night is the gap between what AI systems do and what we can explain about them.

Modern AI models are not programmed in the traditional sense. They are trained. The difference matters because trained systems can produce outputs that are correct without being explainable. A board cannot simply ask, "Why did the system make that decision?" and expect a clear answer.

This is not a technical problem that will be solved soon. It is a fundamental characteristic of the technology.

What I have learned: Boards need to shift their focus from "can we explain every decision?" to "do we have sufficient oversight to detect and correct harmful outcomes?" This is a lower bar, but it is an achievable one. It means investing in monitoring, testing, and human review — not demanding perfect explainability.

The Accountability Problem

The third concern is the most persistent. When an AI system causes harm, the accountability chain is rarely clear.

Is the developer responsible? The company that deployed it? The board that approved it? The customer who used it? The regulator who did not stop it?

The answer is probably all of the above. But that is not how our legal and governance systems work. They expect clear lines of accountability.

What I have learned: Boards must pre-empt this question. Before deploying an AI system, ask: "If this causes harm, who will be held accountable? And are we prepared for that?" If the answer is unclear, the system is not ready for deployment.

What Keeps Me Grounded

Despite these concerns, I remain optimistic about AI governance — not because the problems are easy, but because directors are capable of rising to challenges.

The boards that will navigate this well are not the ones with the most technical expertise. They are the ones that ask the right questions, demand the right information, and maintain the right level of healthy scepticism.

AI governance is not a technology problem. It is a governance problem. And governance is what boards do.

That is what keeps me going — and what keeps me awake, in the best sense.