WHAT A MACHINE MAY DECIDE.

WHAT HAPPENED

The National Committee for the Agentic AI Project ran a workshop for more than 100 federal officials on 9 August. The target behind it came from the Cabinet on 23 April 2026: half of UAE Government sectors, services and operations to run on agentic AI within two years. On 18 May the Cabinet approved an implementation framework for ministries and a training program for 80,000 federal employees, and in June each of 50 federal entities was given 90 days to choose a first service and start building it. Under those numbers sits the part that matters most to private companies: the committee is building a classification for which tasks an autonomous system may execute on its own, and which it may only recommend. Its stated operating principle is "human leads, AI enables."

AGENT PERMISSIONS ARE A BUILD SPEC

Most AI programs we audit have a tool list, a budget and a vendor. Almost none have a written answer to the question that most often blocks deployment: what the agent is allowed to do without asking.

The gap shows up late and costs money. Legal will not sign off because nobody can say what the agent decides. Operations will not hand over the queue because nobody has defined how it fails. The pilot works and everyone is pleased, and then it sits in staging for four months while three departments negotiate what should have been one page written before the build started.

A decision boundary is that page, and it works as a build specification. Every downstream question resolves from it: what gets logged, what triggers escalation, who owns the outcome, what the rollback looks like. Write it first and the integration work becomes ordinary engineering. Write it last and it becomes a committee.

A DECISION BOUNDARY TAKES ONE AFTERNOON

Pick a single workflow and list its steps. Put each step in exactly one of three bands.

BAND	THE AGENT	THE TEST
EXECUTE	acts, logs, moves on	reversible within minutes, and a wrong answer costs less than reviewing every right one
RECOMMEND	drafts, a human approves	the action touches money, a contract, or a customer relationship
REFUSE	stops and escalates	legal, medical, safety, pricing exceptions, and anything with a regulator attached

Two rules make this work. First, an unlisted step defaults to REFUSE, because an agent that improvises on unclassified work is what ends programs. Second, one named person owns each band and reviews the classification on a fixed schedule, because the boundary moves: steps start in RECOMMEND and earn their way into EXECUTE on measured performance.

That last rule is the difference between a policy and a system. Companies that do write a boundary often write it once and never revisit it, so the agent stays a drafting assistant and the return never arrives. Treat the boundary as a ratchet, with evidence behind each click.

PROCUREMENT, LEGAL AND THE COST MODEL RESOLVE FROM ONE PAGE

• Procurement gets a spec. "Which of these steps can your product execute unattended, and can you show us the audit trail?" is a far better vendor question than a feature comparison.
• Legal gets something to approve. Reviewing a one-page boundary is a meeting. Reviewing "AI in customer service" is a quarter.
• You get a cost model. A step only reaches EXECUTE if a wrong answer is cheap and quick to reverse, so a small fast model is usually enough there, provided automated checks and a reviewed log catch its misses. RECOMMEND steps cost reviewer time on every run, so a stronger model that drafts closer to right can work out cheaper. Matching the model to the band is often the largest cost lever we find in an agent deployment.

When the federal version is published, the organizations that wrote their own will already know which bands they got wrong. Update, 21 Oct 2026: we have not found a published version of the classification. In August the committee released a design guide for agentic government services, which asks assistants to get the customer's approval before acting and to hand complex cases to a person, and on 11 October WAM reported that it had approved the project's operating framework.