When an Agentic AI Challenge Extended a Dubai Workshop

After any workshop on agentic AI, one simple question separates content from impact: did participants return to their desks with an interesting idea, or leave with a real problem they were trying to solve? At a Decoding Data Science session at in5 Tech in Dubai on July 16, the available answer appears in a small but verifiable trace. A participant who attended described the session as practical and focused on agent design and loop engineering, then said it marked day one of the Agentic AI Demo Challenge and that he would build an agentic system for a real-world problem over the following weeks.

That does not prove every detail of the announced programme, and it should not be claimed to. It does establish that the workshop took place and connected to subsequent work, a more reliable measure than a registration description alone. In an environment where AI headlines proliferate, an event’s ability to prompt someone to build something testable matters more than the density of terms on its slides.

From conversation to a system that reviews itself
The workshop set out a progression from prompt engineering to context, then to the hosting architecture and loop engineering. The practical meaning of that sequence is not that writing for a machine has lost its value. It is that a product does not become an agent merely because it generates convincing text. It needs knowledge it is permitted to access, specific tools, a workflow, appropriate memory and a way to measure whether it completed or failed at its task.

The distinction matters for business teams in the region. A founder asking a model to prepare customer-service responses, for example, does not need a general promise of autonomy. The founder needs to define what the system can read, when it stops, who reviews a sensitive response and how a case in which it failed is tested again. That is the shift from a demonstration to a process that can withstand daily use.

What remained of the live demonstration
The event page described a case study of an HR assistant using retrieval-augmented generation, tools, workflow and evaluation. The subsequent evidence available, however, does not provide a complete recording or slides that prove every step of the implementation. It is therefore appropriate to discuss the intended model, not its actual performance in the room. What can be known with confidence is that the attendee who documented his participation linked it to a challenge that began that day and said he intended to build an agentic system around a real-world problem.

This is a modest result, but not a marginal one. The challenge moves the learner from receiving concepts about context and loops to taking responsibility for choosing the problem, the data and the success criterion. It also opens a useful space for evaluation: can the system handle a complex case? Does it ask for help when information is incomplete? Can another team understand and review its decisions? These questions reveal the distance between a model that appears intelligent in a short demonstration and a system that genuinely serves an organisation. There is no ready answer, but defining the test before building prevents a team from confusing speed of generation with correctness of outcome.

The measure worth following
Dubai and the region do not lack AI gatherings. What they often lack is documentation of what happened after the gathering. This is a useful starting point for follow-up because the challenge creates time between the workshop and the outcome. If organisers later publish participants’ projects, judging criteria or lessons from failure, the workshop will become a record from which the community can learn, rather than a passing occasion.

For now, the honest conclusion is limited: the practical workshop took place and extended into a build challenge, but the results of that challenge have not yet been published in sources that could be verified. That is precisely what should remain under observation, rather than what a reader may assume from the term agentic AI.