Board Meetings
How AI Is Changing Boardroom Dynamics
Picture a board pack that lands on a Thursday evening. Management's paper on a proposed acquisition runs to forty pages. Clipped to the front is a two-page AI summary: the main risks, a handful of suggested questions and a recommendation to proceed. By Monday's meeting, one director has read only the summary. Another has fed the financials into her own AI tool and come back with a rather different view of the integration costs. A third hasn't opened either and plans to take his lead from the room.
Monday's discussion will be shaped by those three starting points, and nobody is likely to mention it.
Most governance writing on AI (and there is a great deal of it now) is about oversight: how boards should supervise the company's use of the technology. Fair enough. What gets less airtime is the effect AI is having on the board as a working group. PwC's 2025 Annual Corporate Directors Survey found 35% of directors saying their boards had already folded AI into their oversight work. So for roughly a third of boards the tools are already in the room, whether or not anyone has thought much about what they do to the conversation.
This article sits in our series on building AI-ready boards. The other pieces deal mostly with skills and structure. This one stays inside the boardroom, with how directors deal with each other once AI is part of the furniture, and with what a Chair can do about it.
In this article:
How AI influences boardroom interactions
How is AI changing board meetings? Mainly by changing what people walk in with. A director who has spent an evening interrogating the numbers with an AI tool arrives with a different kind of confidence from one who skimmed the summary at breakfast, and a recommendation that "the analysis" supports carries a weight that is hard to pin down but easy to feel. Often the effect is small. Sometimes it decides the outcome.
We'd point to four pressure points. Directors start trusting summaries because they read well, which isn't the same as being right. Participation tilts towards whoever is most at ease with the tools. Challenge gets harder, since there's no obvious author to argue with. And consensus turns up a little too early when the whole board has read the same two pages.
To be fair to the technology, it also fixes something boards have grumbled about for decades. Directors have always seen the business largely through management's lens, and AI gives them a way to check. PwC's analysis on the Harvard Law School Forum notes that directors can now test management's figures for themselves. Stanislav Shekshnia and Valery Yakubovich, who ran focus groups with more than 50 board chairs for HBR, tell the story of a chair who handed four of her board's conclusions to an AI model and asked it to take another look. It backed three. The fourth got reopened, and the second discussion was a good deal better than the first.
So where's the problem? Partly in how quickly faith in these tools is growing. Datasite's 2026 survey of 1,000 dealmakers, carried out with FT Longitude, found 43% saying AI already makes better decisions than humans on some deals. You can argue with that figure. The executives who believe it are still the ones writing board papers.
Partly, too, in how unevenly the knowledge is spread. McKinsey reports that 66% of directors say their boards have limited or no knowledge of AI. On plenty of boards, then, a couple of directors understand the tools properly and the rest are left to defer, whether to those colleagues or to the software.
The Chair's role in AI-assisted discussions
How should Chairs manage AI-assisted board discussions? Treat AI the way you'd treat a clever, persuasive consultant. Listen to it, ask where its numbers came from and be wary of letting it either open the discussion or close it.
Chairs have been here before in other guises. The FRC's guidance on the UK Corporate Governance Code makes the Chair responsible for "setting clear expectations concerning the style and tone of board discussions" and lists a dominant personality or group of directors inhibiting contribution from others among the usual suspects behind bad decisions. The twist now is that the dominant presence might be a document nobody in the room wrote. Or the two directors who know their way round the software.
The agenda is where a Chair has the most say, so that's where we'd start. Compare an item headed "AI market analysis" with one asking "Should we enter this market within 18 months, given the analysis?" The first gets you a presentation and the second gets you a decision. If AI helped produce a paper, the paper should say so and list what it drew on, and the agenda should leave a few minutes for someone to poke at the assumptions. Vincent Bruni-Bossio and his co-authors go further in HBR, warning that careless adoption can end up eroding the Chair's own influence over strategy. A slightly ironic outcome for a tool that was supposed to help.
Trust: how far should directors rely on AI-generated insights?
Only as far as they understand where those insights came from. AI summaries can be wrong while sounding perfectly sensible (PwC's analysis calls these credible-sounding errors), and a summary built from management's own papers inherits management's framing, however even-handed it sounds.
A simple habit helps here. Before an AI-generated view shapes a decision, somebody should be able to say what it was based on, what it couldn't have known and whether its key claims can be checked against something a director could look up. If the room goes quiet at those questions, that tells you something.
Participation: how do Chairs stop AI crowding out quieter directors?
Change the running order. On most boards the first person to speak on an item frames it, and an AI summary that everyone skimmed on the way in can do the framing before anybody has opened their mouth. Chairs can take directors' views before the summary, or management's presentation, gets discussed. They can call first on the director who tends to hold back, or on the one who raised a doubt over coffee. They can also make sure nobody is working from a thinner pack or older tools than their colleagues, because on a board where only two people are comfortable with AI, comfort turns into clout surprisingly fast.
A few boards are experimenting further. INSEAD's write-up of the research describes a Swiss company using AI to measure speaking time and participation in board meetings. Interesting, certainly. We'd only add that directors should know it's happening, since finding out afterwards that the software thinks you talk too much is unlikely to help board relations.
Challenge: how should boards test AI-generated recommendations?
Much as they'd test any recommendation, except that this one can't answer back. Press a CFO on an assumption and you learn something from how she handles it. A model's output just sits there. INSEAD's researchers also found that these tools tend to sidestep contentious debate, and contentious debate is rather the point of a board. The challenge has to come from the directors, then. Some things we've seen work:
- asking management to explain where, and why, it disagrees with the AI analysis;
- getting the tool to make the best case against its own recommendation, which is often the more revealing exercise;
- giving one director the formal job of arguing the opposite view on major decisions, so nobody has to volunteer to be awkward;
- bringing in outside expertise when the stakes warrant it, which FRC guidance has long expected of Chairs anyway.
Consensus: what does good consensus look like with AI in the room?
Agreement the board can explain in its own words, reached after the assumptions have had a proper going-over. When a board agrees quickly around an AI recommendation, it's worth asking why. Maybe the case really is that strong. Or maybe everyone read the same summary and no one fancied being the person who held things up. As Julie Baddeley, Chair, warned on the Boardroom Confidential podcast: “It’s very easy for a board to slip into groupthink where everybody’s reinforcing everybody else’s thinking.” AI-generated recommendations could amplify that risk, particularly when directors rely on the same information without challenging its underlying assumptions.
A Chair can slow things down without making a drama of it: restate the decision in plain terms, ask directly whether anyone has reservations and make sure any dissent ends up in the minutes. Then pin down owners and dates. It feels like housekeeping, but in our view it's the step that turns a nod around the table into a decision the board can stand behind.
Preserving independent judgement
How can boards use AI without losing their own judgement? By using it for preparation and analysis while keeping the decision firmly with the directors. Fiduciary duties belong to people, and Skadden's guidance for directors on the Harvard Law School Forum is clear that handing critical calls to AI without meaningful human involvement could put those duties at risk. It's hard to imagine a regulator being impressed by "the model recommended it".
Where does the line fall? Roughly here. AI is very good at getting through a 200-page pack overnight, noticing a pattern across three years of minutes, suggesting questions and running scenarios. It is far less good at reading whether a chief executive is being candid, balancing a pension scheme's interests against a growth plan or deciding how much risk a company should take on. That second set of tasks is, more or less, what directors are there for.
Secure, human-in-the-loop AI use in the boardroom
Security tends to get forgotten in the excitement. A director who pastes board papers into a public chatbot may have created a real problem, and Skadden points out that AI chats "may be discoverable and could thereby end up in the hands of regulators or an adversary".
We'd suggest boards work to a few plain rules. Board material goes only into secure, approved systems with controlled access. Anything AI-generated is labelled, with its sources visible so directors can check them. A named person reviews AI content before it reaches the pack or the minutes, and the board takes and owns every decision, which is what "human in the loop" ought to mean in a boardroom. McKinsey also recommends ongoing AI education for directors, and we'd agree: it is difficult to oversee something you don't understand.
A quick test for AI-assisted discussions
Three questions a Chair might ask once the meeting is over:
- Could each director explain the decision without leaning on the AI output?
- Did anyone push back on the recommendation, and did the pushback get a proper answer?
- Do the minutes record a board decision, with owners and deadlines, or do they read more like approval of whatever the tool proposed?
A "no" doesn't mean switching the tools off. It does suggest they're thinning out the discussion they were brought in to improve.
How Sherpany supports better board deliberation
Everything above rests on the unglamorous basics of a well-run meeting: a clear agenda, decent preparation, a discussion everyone can join and a proper record afterwards. That's what Sherpany was built for, and our AI Features are there to help at each stage while the board keeps the decisions.
Before the meeting, the Agenda Builder helps Chairs and Company Secretaries give each item a clear objective and a realistic time slot, so the debates that matter aren't squeezed into the last ten minutes. Papers go to every director at once, in one secure place. The Insights Copilot brings in relevant material from earlier meetings, which means the whole board starts from the same context and the AI-confident directors don't get a head start. Directors can also annotate papers and jot down questions in advance, a useful way in for anyone who finds it harder to interrupt.
Afterwards, the AI Minutes drafts structured minutes from the discussion, picking out decisions and tasks for people to review and approve. Actions are tracked with owners and deadlines, and votes and approvals run through an auditable process. The drafting is automated; the decisions stay with the board.
AI should sharpen the debate it joins
Much of what AI brings to the boardroom is good news: better-prepared directors, less reliance on management's version of events and, with a bit of luck, more meeting time for the questions that deserve it.
None of it arrives automatically. Chairs will need to keep an eye on who trusts what, who is talking and who isn't, whether anyone is still prepared to disagree and what a show of agreement really means. Get that right and AI should make board debate sharper. Neglect it, and meetings may well get shorter and smoother while the decisions coming out of them become harder to defend. We'd much rather see AI strengthen human debate, challenge and judgement than stand in for them.
If you'd like to see how AI can support board deliberation with directors firmly in control, book a free demo of Sherpany and we'll show you how it works.