The conversation about AI in the workplace has been dominated by the wrong question.
Boards ask: how many roles can we eliminate? Consultants model headcount reduction scenarios. Vendors promise automation rates that sound impressive until you ask what they are actually measuring. And the organisation spends eighteen months in a low-grade state of anxiety while nothing material changes.
The question that actually matters is different: how do we redesign the way work gets done so that the people we have can achieve outcomes that were previously impossible?
That is an operating model question. Most organisations are not asking it.
What AI actually does to work
AI does not eliminate jobs cleanly. It eliminates tasks - specifically, the low-judgement, high-volume tasks that sit inside every role and consume the time that should be going to higher-order work.
A customer service agent spends roughly forty percent of their time on wrap-up: summarising calls, updating records, categorising cases. AI can do that. The agent's remaining sixty percent - de-escalating an upset customer, exercising discretion on a borderline case, building the kind of rapport that retains a long-term account - AI cannot do that reliably, and is some distance from doing so.
If you automate the forty percent and leave the operating model unchanged, you have a slightly faster version of what you had before. You have not actually changed what the agent is capable of delivering.
If you redesign the role around the sixty percent - with AI handling the rest as a matter of course - you have something fundamentally different. Agents handling more complex interactions. Resolution rates improving. Customer satisfaction following. The same number of people, delivering materially more value.
That is augmentation. It requires an operating model built around it, not a pilot bolted onto the existing one.
Why operating models lag
Operating models are slow to change for understandable reasons. Roles, responsibilities, governance structures, and performance frameworks are all interconnected. Changing one without the others creates confusion. So organisations tend not to change any of them until they absolutely have to.
The result is that AI arrives into operating models designed for manual processes. Approval chains built for human review of every decision. Reporting lines organised around activities, not outcomes. Job descriptions that list tasks rather than accountabilities. Performance metrics that count volume rather than value.
Into this structure, AI is introduced. And the organisation discovers that the AI is faster than the process around it. Decisions still queue. Escalations still take days. Customers still wait. The AI did its part; the operating model absorbed the efficiency and produced nothing.
This is not a technology problem. It is a design problem.
The GCC has an advantage here
Across the UAE, Saudi Arabia, and the wider Gulf, there is a genuine structural opportunity that established markets do not have.
Organisations building new functions - new shared service centres, new government digital services, new financial institutions - are not inheriting legacy operating models. They are designing from scratch. The question of how AI fits into the operating model does not require dismantling something that already exists. It can be answered before the operating model is built.
The UAE's national AI strategy and Saudi Arabia's Vision 2030 programmes are accelerating this. Public sector entities are being asked to demonstrate AI-enabled service delivery, not just to deploy AI as a technology curiosity. That pressure creates an incentive to design operating models that actually work with AI rather than despite it.
For international organisations entering the GCC, the same logic applies. Do not replicate your home-market operating model and add AI on top. Design for the region with AI as a first-order consideration. The organisations that do this now will be ahead of their competitors in three years - including their competitors in markets that have been talking about AI longer.
What the redesign looks like in practice
Redefine roles around judgement, not tasks
If a job description lists tasks that AI will perform, it is already out of date. Roles need to be redefined around the judgements only a human can make: what to do when the AI recommendation is wrong, how to handle cases that fall outside the model's training, when to override the system and why. These are not lesser functions. They are the functions that determine whether the AI programme actually delivers value.
Remove approval layers that no longer add value
Many governance structures exist because manual processes required human checkpoints at every stage. When AI is performing those stages, the checkpoint is often redundant. Leaving it in place means you have automated the work but not the delay. Identify which approvals exist to catch genuine risk and which exist because that is how it has always been done. Remove the latter.
Change what you measure
Volume metrics - calls handled, tickets closed, documents processed - made sense when volume was the constraint. When AI removes the volume constraint, measuring volume tells you nothing useful. The metrics that matter are outcomes: resolution quality, customer effort, decision accuracy, time to value. Operating models that still report on volume will underestimate what AI is delivering and over-invest in things that no longer create competitive advantage.
Invest in the human skills that AI amplifies
Augmentation only works if people know how to use what they have been given. That means training - not on the AI tool itself, but on the higher-order work the tool is making space for. Judgement. Communication. Complex problem-solving. Emotional intelligence in high-stakes interactions. These are not soft skills. They are the skills that determine whether an AI-augmented workforce outperforms an unaugmented one, or just does the same things more cheaply.
The bottom line
The companies that win with AI will not be the ones that automated the most headcount. They will be the ones that redesigned their operating models so that the people they kept could do things that were previously out of reach.
That requires treating AI adoption as an organisational design exercise, not a technology deployment. It requires changing roles, governance, metrics, and capability investment at the same time as changing the technology stack.
It is harder than buying a platform and watching the pilot results come in. It is also the only approach that actually delivers on what AI promises.
Your people are not the bottleneck. The operating model that has not changed to make use of them is.