The Board Signed the Contract. It Did Not Understand the Decision.

Boards are approving massive AI contracts without grasping the shape of the transformation they have just commissioned — and often without realising a transformation is what they commissioned at all. The approval is the thing that will be held against them.

Picture a board meeting. A normal one.

There is a paper in the pack. It proposes a contract with an AI provider — a large one, a name everyone in the room knows. The cost is set out. The efficiency case is strong. Two of the big consulting firms have looked at it and support it. There are questions, good ones, about price and timing. The vote is carried. Everyone moves to the next item.

Nothing about that meeting was careless. The process worked exactly as it should.

And not one person at that table could have told you what they had just let loose inside the business — or the scale of the potential liability they had signed for.

That is what this piece is about. Not AI arriving in secret. AI arriving in plain sight, with a signature on it — approved, minuted, filed — and a board that still cannot say what it agreed to.

What the board actually bought

The paper said “software”. It wasn’t, quite.

The thing being sold now is not a piece of software, run by a person who is answerable for it. It is closer to a worker. You give it a task and it carries the task out on its own — it plans, it acts, it moves between systems, it changes what it does as it goes. The industry has a word for it. It calls it an agent.

And the people selling it have stopped using the language of software altogether. They call it a digital workforce. They talk about deploying not a system but a team of these agents — dozens at a time, each one doing a job.

In plain terms: an army of software robots working inside the company’s systems, at a scale and complexity most boards have no real way to picture, let alone govern.

That language is doing quiet work. A workforce is something a board expects to be managed — by someone else, further down the organisation. It is not something a board expects to sit and govern itself, line by line. So the words alone move the decision off the board’s desk at the very moment the board is signing for it.

The providers have made saying yes easy. The agents are sold in ready-made bundles now — a pack for finance, a pack for customer service — switched on together rather than built one at a time. And the AI companies and the big consultancies increasingly sell it side by side, as a single offer. What arrives on the board table is not a product the board can picture. It is a capability the board has been gently encouraged not to look at too closely.

Why the signature is not the safeguard

A board can approve something it does not understand. It happens all the time.

It happens whenever the paper explains the cost and the benefit but not the machine underneath. The board sees what it will pay and what it should get back. It does not see what the thing can actually do once it is inside the business and running.

With an agent, that gap is wide, and every part of it is a risk the board is now carrying:

• What data each agent can reach and read.
• What it is allowed to do without asking a human first.
• How the list of things it is allowed to do will quietly grow over time — often without anyone deciding that it should.
• And who is accountable when one of these agents acts in a way no one expected, and creates a loss, a breach, or damage the company must then answer for.

Those are the governance questions. The signature answers none of them. It settles who supplies the software. It says nothing about who is accountable when the software acts — and acts in a way the board never saw coming.

Then something predictable begins to happen, and it begins from a worse place than usual, because the board believes this ground is already cleared.

The agent’s authority widens a small step at a time — the gradual drift in which drafting becomes sending and a human check becomes an approval no one reads. Not one of those steps comes back to the board. And the whole time, the organisation believes a human is still in control — believes it more firmly than it should, because there is a signed contract in the drawer that says the matter was dealt with.

The people who advised you may be paid if you say yes

There is one more turn, and it is the one boards have not yet noticed.

For a hundred years, the consultant billed for time. Advice in, invoice out, and the consultant then walked away, with no stake in what you chose to do. That is changing. The big firms are moving to being paid on results — a share of the saving, a cut of the outcome the AI is supposed to deliver. And those same firms often hold formal partnerships with the very AI companies whose agents the board is about to buy.

So look at the room again. The board takes its comfort — its sense that clever, independent people have checked this — from an advisor who may now earn more if the answer is yes, recommending a system that advisor is partnered to sell.

No one here is acting in bad faith. That is not the point. The point is a simple question the board should be asking and usually is not: if the people advising us are paid more when we say yes, could their advice be compromised? The advice may well be sound. But an advisor who profits from the decision is no longer a neutral judge of whether you should make it — and the board has taken a sales relationship for an independent one without noticing the difference.

Why this cannot wait

The exposure is not somewhere in the future. It begins the day the contract is signed.

The capability is real from that moment. So is the risk. And if something goes wrong — information exposed, a bad decision made at machine speed, a customer or a patient harmed — the company will be asked one question. Show us how you governed this.

Here is the trap. The board’s own approval — the paper it felt reassured by, the vote it carried so smoothly — is the document that gets produced in answer. Not as proof that the board was in control. As proof that the board saw the decision, signed it, and did not understand it.

That is the danger in a single line. The signature the board took comfort from is the signature that will be used against it.

Governance was never a signature on a contract. It is the board being able to say, before it signs, in its own words and not the salesman’s, what it is agreeing to. What these agents may do. What they may reach. How far they may go on their own. Where a real human being still stands in the way. Who answers when one of them acts.

If the board cannot say those things plainly, it has not governed the decision.