Thirty-Two Advisors in the Cabinet Room
Co-Founder & CEO of CLRT
Two announcements landed a day apart at the start of September, and read together they describe the whole distance between the state and the firms it is asking to follow. On 1 September, Dubai Chambers launched training tracks aimed at more than 14,000 member companies, delivered through a new e-learning platform, with 90 chairpersons of business groups and councils in the room. On 2 September, the UAE Cabinet approved a system of 32 specialised AI advisors that analyse policies and legislation, assess their impact and put forward recommendations, working with ministers 24 hours a day, alongside a governed framework for issuing Cabinet decisions with agentic support. One side of the economy was handed a course. The other side put agents in the room where the decisions are made, and published its scorecard on the way in.
Start with what the Cabinet release actually contains, because the advisors are the headline and the numbers are the point. The same sitting reported the federal government's 2025 results: more than 308 million digital transactions across 1,633 digital services, 88.5 million visits to government websites, 15.8 million application downloads, 471 digital transformation projects delivered, and customer satisfaction with priority services at 92 percent. Abu Dhabi had already gone further in January. Its Department of Government Enablement reported that the TAMM platform resolves 95 percent of requests through AI, completed more than 1.9 million service cases through conversational AI, holds a 92.5 percent satisfaction rate and eliminated more than 36 million customer visits a year. Those are the operator's own figures and should be read as such. But notice their shape. Volume, resolution, satisfaction. The state has told the private sector, in the most public way available, exactly which three numbers it considers the measure of an agentic operation.
Now look at what the private sector received in the same week. The Dubai Chambers programme is a training platform: specialised agentic AI tracks for the business groups and councils under the Chamber, with incubators and an executive committee behind it. It sits inside the wider two-year plan announced in May. Nothing in either announcement is wrong, and a chamber of commerce cannot install an agent inside a member's back office. But the sequence matters. Abu Dhabi trained 95 percent of its public-sector employees, put more than 1.9 million service cases through conversational AI on TAMM, and published resolution and satisfaction figures for the flagship. Training, deployment, measurement. The private sector has been handed the first of those three and is treating it as the whole assignment.
The private firm's real disadvantage is not that the state has more money or better engineers, though it has both. It is that the state instrumented itself before it automated itself. A resolution rate of 95 percent means someone defined what a resolved request is, counted every request for a year, and had the confidence to print the result. A 92 percent satisfaction score means the outcome was checked with the person who received it, not inferred from a dashboard. Very few UAE companies could produce either figure for a single workflow this afternoon. They know their revenue and their headcount. They do not know how many customer requests entered the business last quarter, what fraction were closed without a second touch, or how the closed ones were received. Without that baseline, an agent cannot be pointed with any precision, and its results cannot be distinguished from noise.
This is where the training course becomes a trap rather than a start. A firm that trains its people and then deploys an agent into an unmeasured workflow will get a number afterwards, because every vendor supplies one. The number will describe the agent's activity, not the firm's outcome, and it will be the only number in the building, so it will be believed. The state's advisors are not doing that kind of work. Thirty-two of them are assessing the financial, economic, social and environmental impact of each subject that reaches the Cabinet, which is to say they are pointed at the highest-value, highest-judgment step in the entire government and are held to a framework that names confidentiality and cybersecurity as its principles. The contrast with a chatbot in a customer inbox is not one of ambition. It is one of knowing which decision is worth an agent's attention and being able to say afterwards whether it improved.
Read the two announcements as a single instruction and the instruction is uncomfortable. The state is not asking firms to learn about agents. It is telling them the terms on which agentic operations will be judged, by publishing its own report card first: how many transactions, how many resolved by the machine, how satisfied the people on the other end were. Any company that takes the mandate seriously should assume those are the columns it will eventually be asked to fill, by a bank, a regulator, an acquirer or a large customer that has read the same releases. The firms that will look competent in two years are not the ones with the most certificates from the Academy. They are the ones that can state their own three numbers today, before the agent, and can therefore prove what it changed.
The state instrumented itself before it automated itself. The private sector is being trained before it has measured anything.
A deeper dive
The mechanism worth understanding is why the state could publish these figures and a private firm usually cannot. A government service is a transaction with a defined start, a defined end and a citizen who can be asked how it went, which is why a platform like TAMM can count 1.9 million conversational cases and report 95 percent resolved through AI. The private firm's workflows are rarely defined that cleanly. A customer request arrives by email, is forwarded twice, is partly answered on a call and partly in a spreadsheet, and is closed in someone's head. There is no event that marks the end, so there is nothing to count, so there is no rate. The work of making a workflow measurable is unglamorous: deciding what counts as a request, what counts as resolved, where the timestamp lives, and who owns the satisfaction question. It is also the work that determines whether an agent can be pointed at the workflow at all, because an agent needs the same definitions the measurement does. The state did that work in the years before it announced agents. Most firms are being asked to skip it.
The second-order trap is that the vendor's number arrives to fill the vacuum. Every agent platform reports what its agent did: conversations handled, tickets touched, hours saved by an assumed multiplier. Those figures are real, but they measure the agent's activity inside a workflow whose baseline was never taken, which means the before and after are not comparable and the gain is unfalsifiable. A firm in that position will publish the vendor's figure because it is the only one available, will build the next year's budget on it, and will discover the gap when someone with a claim on the business asks the state's question in the state's terms: what fraction of your requests does the machine resolve, and how do the people on the receiving end rate the result. The Abu Dhabi figures are self-reported by the operator and deserve the same scepticism as any operator's figures. The difference is that the operator defined the metric, counted for a year and published it with its name attached. That is the standard. A private firm that cannot meet it will not be judged on its training hours.
Work with CLRT
Defining the baseline before the agent is the work CLRT does. We take the one workflow that matters, decide what a request and a resolution mean in your business, put the timestamps and the satisfaction question where they belong, and only then decide where an agent should be pointed and what it must move. If you want to know which of your workflows could stand up to the state's three questions, and what an agent would be worth there in dirhams, CLRT Ascent runs that diagnostic. Bring us the workflow you would be embarrassed to be measured on. That is usually the one worth measuring first.
Vishal Sachar is the Co-Founder and CEO of CLRT, where he helps UAE businesses make sense of applied agentic AI and put it to work. He writes on agentic systems, AI governance, and the economics of automation. Reach him at vishal@clrtstudio.com or on LinkedIn.
