How AI agents are actually used across MENA businesses today

In production, an AI agent does a defined operational job, not everything at once. The patterns Sheba runs most often are: answer customer questions in Arabic across the channels a business already uses, qualify inbound leads before they reach a person, take orders and bookings, and hand off to a human the moment a conversation needs one. Each agent is scoped to a real workflow with a clear definition of done, so it does that job consistently rather than improvising. This is the difference between an operational agent and a chatbot demo — the agent connects to the business's data and processes, and the work it completes is the same work a team would otherwise do by hand.

An inbound message is the start of a workflow, not just a question. For lead qualification, the agent asks the qualifying questions a sales team would ask, captures the answers against the customer record in the CRM, and routes only the ready buyers to a person — so the team spends its time on conversations that are worth a human. For order and booking intake, the agent collects the details a transaction needs, confirms them back to the customer, and records a structured entry the operations team can act on. Both patterns run on the same channels customers already message on, so there is no new app for anyone to adopt.

Across roughly two years of shipping, Sheba runs 35+ AI agent personas in production across 14+ industries — banking and microfinance, healthcare, education, retail, manufacturing, food and beverage, real estate, travel and documentation, and more. Each persona carries the vocabulary, tone, and rules of its industry rather than one generic script, because a banking conversation and an F&B order are not the same job. The breadth comes from the same operational patterns applied to different workflows — and where a business needs something beyond the four standard channels or the standard support flow, that work goes to custom AI.