Where a board should start with AI

The first move for almost any board that decides to take AI seriously is naming a committee. It is also the move that resolves the least: it relocates the question instead of answering it.

Carlos Andrés Ramírez · 2026-08-28

The first move for almost any board that decides to take AI seriously is naming a committee. It is also the move that resolves the least: it relocates the question instead of answering it.

I have sat in the first meeting where a company decides to get serious about artificial intelligence, and the pattern barely varies. Someone proposes a committee, everyone nods, and within three weeks there is a monthly calendar invite, an agenda owner and a deck carrying the initiative's own logo. It looks like a start. It isn't. It is a comfortable way to postpone the four decisions that actually are.

Naming a committee costs nothing politically. Nobody objects to a room where AI gets discussed. What does cost something, and what gets avoided because of that, is deciding who answers for the result on the P&L, what happens to a pilot once it has been alive for six months, and under what criteria an initiative gets killed before it becomes impossible to close. A committee without those three answers in writing is not a starting point. It is a waiting room with minutes.

The symptom

Where should a board actually start with AI?

The answer circulating today, copied across dozens of vendor proposals sized for smaller companies, is a three-step recipe: form a committee, write principles, pick two pilots. It works for a company running its first experiment. It does not work for a board that already has ten initiatives in flight and needs to know, two quarters from now, which ones go live and which ones get shut down. The recipe solves the launch and goes quiet exactly where the hard part starts.

• The committee meets monthly and reviews the same slides, almost always the same two pilots described as progressing.
• Nobody in the room can say, without opening a spreadsheet, how much the initiative has cost so far.
• Pilots carry no decision date; they get reviewed indefinitely until someone stops asking about them.
• The committee recommends and the full board never votes against it. In practice, it ratifies.
• Ask who loses their bonus if the pilot never reaches production, and the room goes quiet.

The problem underneath

A committee that only recommends decides nothing, and everyone in the room knows it.

In most boardrooms, the AI committee looks like the innovation committee from a decade ago: a space for visibility with no authority to spend or to close. It recommends, but it does not sign off. It reviews, but it cannot kill an initiative without escalating to the full board, and escalating something to the full board purely to kill it burns enough political capital that it almost never happens. The result is predictable: initiatives do not die, they accumulate. And a board with fifteen initiatives sitting in accumulation does not have an AI portfolio. It has a graveyard that still makes payroll.

The cause is not that the committee is badly staffed. It is that it was asked to answer a corporate governance question (who decides, with what authority, with what consequence) using a coordination tool. A committee coordinates well. Deciding is a different competency, and only the person who stands to lose something for deciding wrong has it: budget, bonus, credibility in front of the board.

A committee that only recommends does not govern AI. It gives the company the feeling of governing it.

BECOME

The framework

What does a board need to decide before naming a committee?

Mandate

What the committee can decide without escalating to the full board, and what it cannot. If the answer is 'nothing, everything gets escalated,' there is no committee. There is a status meeting.

Result owner

The person, not the committee, who answers for each initiative's number on the P&L. A collegiate body never loses a bonus. A named person does.

The six-month test

The date by which every pilot has to do one of two things: move into operation with its own budget, or shut down. Without that date in writing, a pilot lives indefinitely in demo mode.

Kill criteria

The signals that are enough to kill an initiative before it turns into a sunk cost nobody wants to be the one to close. If the only kill criterion is 'once it clearly no longer works, even dressed up,' there is no criterion.

Reporting cadence

How often the committee reports to the full board, and with what evidence beyond last month's same slides. Without that cadence, the committee answers only to itself.

None of the five require understanding a model. They require the same discipline a board already applies to an acquisition or a cost-cutting plan: name who answers for it, put a date on the decision, and accept that part of what gets approved today will close tomorrow without that being a failure. When those five answers exist, forming the committee is a one-day formality. When they do not, the committee substitutes for having made them, and it lasts for years.

Before calling the committee's first meeting, name who answers for the result of the one initiative already running. If nobody on the board can name that person, you do not have a committee left to form. You have a governance decision left to make, and forming the committee will only postpone it another quarter.

Let's design your AI committee's mandate