News

Hilda Maalouf, Oxford-certified AI expert and author, advising on AI strategy, governance and institutional transformation.

gettyKnowledge is becoming free; judgment is becoming priceless. I made that case in my last article, and the question that follows is harder: What happens to judgment when software stops advising us and starts acting for us?

This is no longer a thought experiment.

Deloitte’s 2026 “State of AI in the Enterprise” survey of more than 3,200 business and IT leaders across 24 countries found that 23% of companies are already using agentic AI at least moderately today. However, that number is expected to rise to 74% within two years.

Yet the same survey found that only 21% of companies have a mature governance model for autonomous agents. Gartner expects that more than 40% of agentic AI projects will be canceled by the end of 2027, citing “escalating costs, unclear business value or inadequate risk controls.” Put simply, we are delegating authority to machines faster than we are deciding who answers for what they do.

The Delegation Problem We Already KnowNone of us would hire a capable new vice president and say, “Do whatever seems right.” We would define their mandate, their limits, what they escalate and whom they report to, and we would expand their authority only as their judgment proved itself. That is what a chain of command is at its best; it’s not bureaucracy, but the way institutions scale judgment beyond the few people at the top.

Agents are a new kind of delegate. They draft, reconcile, route, negotiate and increasingly transact. Yet many are still deployed as tools rather than delegates: They’re switched on, connected to systems and left to run. When something goes wrong, “Who approved this?” too often has no answer.

Constitution Before CodeWhen I began building my own AI chief of staff this year, I made a decision that slowed the project down and turned out to matter most. Before writing a single routing rule, I wrote the system a constitution. It defines what the agent may decide alone and when it must pause or stop, as well as the limits of its authority and how it behaves under uncertainty. The directive at the top was deliberate: Optimize not for productivity, but for strategic value, coherence and the protection of my attention. The technology was the easy part. Agreeing on those rules in plain language was the real work.

I now believe every agent needs a constitution, and that three articles matter most:

1. Mandate: A written scope of what the agent may decide, what it may only recommend and what it must never touch. A mandate that lives only in a developer’s configuration file is not a mandate. It should be readable by the business owner who will answer for it.

2. Escalation: The most valuable design choice is often the moment an agent stops and asks. Thresholds based on value, risk, novelty or customer impact should route the decision to a named human. In my own system, agents prepare; an approval queue decides what leaves the building.

3. Accountability: Every agent needs a human owner, just as every budget does, and an audit trail that shows what it did, not only what it produced. If we cannot reconstruct why an agent acted, we cannot defend that decision to a regulator, a board or a client.

A Leadership Question, Not An IT QuestionIt is tempting to hand agent governance to our technology teams. I would argue it is the one delegation we should resist. In my Three Layers of AI-Augmented Leadership framework, the first layer is technical fluency: understanding enough about how these systems reason and fail to know exactly what we are delegating.

The second layer is ethical stewardship: the leader’s responsibility for how AI decisions affect people, trust and reputation. Deciding what an agent may do on a customer’s behalf is not a technical setting. It is a statement of values, risk appetite and brand promise, and those belong at the executive table.

The third layer, human-centered communication, matters just as much. Employees need to know which decisions are now shared with machines and where their judgment remains decisive. Customers deserve to know when they are dealing with an agent and how to reach a person.

Why This Matters More In The Arab RegionFor those of us leading institutions in the Arab region, the stakes are higher. Deloitte’s 2026 Middle East findings from its 2026 “State of AI in the Enterprise” report show that 84% of organizations in the region haven’t redesigned jobs or workflows around AI yet, while 77% of organizations consider a vendor’s location important when choosing AI platforms. We are about to hand decisions to agents inside workflows never redesigned for them, even as we care deeply about where our AI lives and whose rules it follows.

Many of our institutions are also led by founding families and close-knit boards, where authority has traditionally been personal rather than written. Agents cannot work that way. They need authority made explicit, and that may prove to be one of the healthiest governance disciplines our region adopts.

Autonomy Should Be Earned​Deloitte’s research suggests (registration required) that the companies getting agents right are the ones resisting the rush to full autonomy. They let agents prove themselves first on work where mistakes are cheap, and they write the rules while the agents are learning, not after.

That mirrors how we develop people. An agent can begin by recommending, move to acting with approval and only then act alone within defined limits. Projects fail less often when their scope matches the trust they have earned.

The Question For Our BoardsAs agents multiply, the defining capability of leadership will not be how many decisions we automate, but how deliberately we distribute judgment between people and machines. Knowledge is free. Judgment is priceless. And judgment at scale requires a constitution.

So perhaps the question for your next board meeting is not “How many agents are we running?” but “Could we name, for every one of them, the person it answers to?”​

Forbes Business Development Council is an invitation-only community for sales and biz dev executives. Do I qualify?