Short answer. Dubai Chambers has opened free agentic AI training to member companies, and the course is worth taking. But 14,000 is the size of the membership rather than the number of sign-ups, and what a course leaves behind is people who can use the tool. Work changes once a process runs on its own, without somebody opening a chat with a model every time.
A course always finishes before the work does.
What was announced?
Free agentic AI training for member companies of the chamber. It opened on 1 September.
Teaching runs through Dubai Chambers Academy, a new e-learning platform, and the tracks cover operations, efficiency, productivity and business decisions. The chamber showed them to ninety heads of its business groups and councils.
This is not a standalone move. In May the Crown Prince of Dubai, Sheikh Hamdan bin Mohammed, announced a two-year programme with one goal: make the city the world's leading place for commercial adoption of agents by 2028. The programme promised four things: training, incubators, opportunities for young people and dedicated funds. Training surfaced first.
The other three sit in the future tense with no sums attached, which is normal for a programme with eighteen months left to run. Reading it as something that has already happened is premature.
Where does 14,000 come from?
It is the membership of the business groups and councils under Dubai Chamber of Commerce. How many companies signed up, the chamber has not said.
The difference is simple. 14,000 shows how many companies can take the training. How many did, nobody has counted out loud. In the retellings the two numbers merge into one, and the picture that comes out has all fourteen thousand already studying.
None of this is a complaint about the chamber. It announced the reach of the programme, and the figure is honest. What matters is elsewhere. If you are one of those companies, you are not yet among the trained. You have an open door and nothing more.
What does the course not give you?
A process that runs when nobody opens a chat with a model.
After a good course a company has people who can frame a task for the model and can see where it invents things. That is valuable. The work itself does not change yet: enquiries move the way they always moved, except now one of the managers sometimes asks a model for help with a reply.
One question at a Monday meeting a month later settles it. What do we do differently from August? If the answer names people who "started using it", the work stayed the same. If it names a step of a process that now runs without a person, the work changed.
In June 2025 Gartner published a forecast: more than 40 per cent of agentic AI projects will be cancelled by the end of 2027. The reasons the analysts give are organisational. Rising costs, unclear value, weak risk controls. A course solves none of the three, and that is the part worth taking away.
The same release carries a second number, useful when you pick a supplier. Of the thousands of vendors selling "agents", Gartner counts about 130 as real. The rest is a fresh label on chatbots and automation scripts written long before any of this.
Which process should you start with?
A process that is frequent, describable, and has a known cost of error. Three signs instead of intuition.
Frequent means the process repeats at least a few times a week. With a rare process you will see no result and collect no examples that show the system behaving correctly.
Describable means the process fits on one page as a series of steps. If it does not fit, the model is not the problem: you have no process yet, you have the habit of a few people.
Known cost of error means you know what happens when the system gets it wrong. "Nothing terrible" will not do here. A specific answer will: the client gets the wrong price, and we fix it by hand within an hour.
| Process | Frequency | Fits on a page | Cost of error | Take it first |
|---|---|---|---|---|
| Replying to an enquiry | every day | yes | wrong reply, fixed in minutes | yes |
| Preparing a quote | a few times a week | partly | wrong price in a document | yes, with review |
| Screening CVs | in waves | yes | a good candidate dropped | yes, if the flow exists |
| Reading contracts | twice a month | yes | a missed clause, expensive | later |
| Monthly accounts | once a month | yes | wrong figures, expensive | later |
The last column sets an order of work, and nothing on it is off the table for good. Contracts and accounts can go to a system too, just later than the one you assemble in the first month.
One honest decision point. If the job is ordinary and a packaged product solves it, buy the package: it will be faster and cheaper. [Building your own](/custom- development) earns its place where the process is what makes you different. Check separately whether you need an agent at all, because a plain workflow handles some jobs the same way every time and costs less to run.
What can you do in the week after the course?
Pick one process, write down how it runs today, and give it an owner.
- Pick one process using the three criteria. One. Nobody finishes three in parallel.
- Write down how it runs today. Who does what, in what order, where people wait. One page, no polish.
- Mark the step where the decision happens. That step takes five minutes, while the rest of the time is spent collecting paperwork around it.
- Name an owner. A person. A department is never an owner. They answer for what the system does, and they are the one who switches it off.
- Agree on what "working" means. One number, measured before launch. Without it, the argument a month later is about impressions.
No agent will be running by Friday, and that is fine. What you get is a page describing the process, a name, and one number. That is enough to talk to a supplier about specifics, and the sentence "we need AI" stops being necessary.
Take the course anyway. Not because a system appears afterwards, but because your people stop waiting for magic and start asking where the system gets its data. In a company where those questions get asked, a system gets built.
Danil Ivanov
Founder, KAIVIX
Builds AI systems for companies in the UAE and beyond.