Industry Insights
|
November 19, 2026
|
-
By Toan Huynh, Board Chair

Leading at the Board Level in the Age of AI

Stop asking if management is ready for AI—ask if your board is equipped to govern it. Toan Huynh outlines the five questions directors must ask to close the oversight gap, and why honest candor beats polished AI presentations.
Overview

Leading at the Board Level in the Age of AI

Before asking whether management is ready for AI, boards should ask a question too few boardrooms ask: are we equipped to govern the decisions we are authorizing? The exposure today is not just technological, it is the widening gap between the people deploying AI and the people overseeing it. That is where risk, liability, and skills shortages concentrate.

I've served on public, private, and PE-backed boards, and I think hard about how to steward decisions when I don't have all the answers. Here is the uncomfortable truth about this wave of AI: none of us do. Not management, not employees, not even the firms building frontier models. So instead of hunting for answers, let's start by asking better questions, and challenging how each of us approaches the technology. I developed the ones below with a group of seasoned directors.

The proof is out there

The best-run AI programs are already publishing their results, and they reward careful reading. Klarna's AI assistant handled 2.3 million customer conversations in its first month, two-thirds of its service chats, the equivalent work of 700 full-time agents. Impressive. But fifteen months later, its CEO admitted cost had been "a too predominant evaluation factor" and the company began hiring humans again, the AI stayed, but the target it optimized for had been wrong. That is a governance lesson wearing a case-study costume: the question was never whether the technology worked, but whether anyone asked what the freed-up people should do next.

Contrast Moderna, which deployed ChatGPT Enterprise to thousands of employees and saw 750 custom GPTs built in two months. Its Dose ID assistant analyzes clinical data to support dose selection, but it provides rationale, cites sources, and defers to a human team that makes the final call. The humans have the competence, information, and authority to overrule the machine. That is what a real control looks like.

Five questions directors should ask

If a director could ask management only five questions:

  1. Which strategic outcome justifies this use of AI, and how will we measure improvement against today's approach?
  2. Where is AI actually being used, including outside approved programs, and who owns the outcomes?
  3. What could go wrong for customers, employees, and the business, and what evidence shows the controls work?
  4. If a critical AI system failed tomorrow, who could stop it, what replaces it, and when was that response last tested?
  5. What capability gaps exist across management, the workforce, and this board, and how will we close them?

Question four is the sharpest test of a plan versus a presentation. Only an answer with named owners, tested fallbacks, and rehearsed escalation proves anything. And if AI cannot be traced to strategy, that is a finding, not a reason to let it live as a side-show.

Old principles, new terrain

Accountability, independent challenge, proportionate controls, these still hold. The wrong instinct is assuming institutional approval completes oversight. Frontier risk gets the headlines; the more probable event is a deployed system failing quietly, at scale. Practice makes perfect, the more tabletop exercises the better. I still remember running phone-tree disaster recovery drills every Friday at a global bank. Boring then, invaluable now.

Boards should also hear from the frontline before hearing the polished version. Employees see the workarounds and the extra verification quietly added to their day. Distinguish task absorption from job elimination, and demand credible transition plans rather than blanket reassurance. And treat "human in the loop" skeptically: a human checkpoint is not a control unless the human has the competence, time, information, and authority to stop the decision, as Moderna's clinical teams do.

Board composition comes first

The question is not whether to add an AI-domain director; it is whether the board as a whole can govern AI. Practical steps: assess the board's skills against the company's actual AI exposure (not generic "digital literacy"); use the systems yourself and see where they fail; review real workflows, not vendor demos; rehearse failure scenarios with management; and hear regularly from frontline employees and independent experts. Embed AI oversight in every committee's remit, risk, audit, compensation, strategy, because distributing oversight must not diffuse responsibility.

Judgment under uncertainty

None of us were trained for this moment. Under uncertainty: identify the unknowns that could change the decision, seek the expertise, and set conditions for proceeding, pausing, or reversing. Delegate analysis, not accountability. Build a culture where every director can say "I don't know", candor should invite inquiry, not undermine credibility. And challenge the old habits, groupthink, heritage thinking, clubby deference, before they turn uncertainty into false consensus. Boards that cannot say "I don't know" will get AI wrong.

CLOSING THOUGHT

The boards that thrive will not be those with the most impressive AI presentations, but those honest enough to govern what they do not yet fully understand, a rare combination of humility, curiosity, and strategic thinking.

Ready to see this framework in action?

Connect with a Naitiv architect for a 30-minute walk-through of how we apply AI governance principles to real ServiceNow programs.
Connect With Us