// Straight answer
Short answer
Usually both, with different jobs. An AI agent suits after-hours replies, the first answer, routine questions, qualifying and booking. A person suits complaints, judgment calls, negotiation and closing. Decide by listing tasks and hours, not by comparing salaries, and build in a clear handover from agent to human.
Answered by Tarek Abou Rjeily, AI consultant & builder in Beirut, for businesses in Lebanon and the Gulf.
The honest comparison is not agent against person. It is which tasks need a person and which do not, across which hours. A receptionist covers a shift and brings judgment and a human voice. An agent covers every hour but only the tasks you have defined.
This page gives no salary figures for Lebanon or the Gulf, because wages differ widely by role and city and there is no reliable public source to cite. Take your own cost for the role and compare it with the hours it leaves uncovered and the messages it cannot get to in time.
An agent is strongest on messages that follow a pattern. It can reply within seconds at night and on weekends, answer questions from a fixed set of facts, ask the qualifying questions, offer booking slots and write the result into a record. It does not tire, forget or get impatient on the fortieth question about opening hours.
A person should own anything where the cost of a wrong answer is high or the situation is new. That includes complaints, refunds, price negotiation, sensitive topics, regulars who expect a familiar voice, and the final step of closing a deal. People also notice what is not said, which an agent will not.
The design rule is simple: the agent never pretends to be a person and never traps a customer. It should say it is an automated assistant, hand over on request or when it is unsure, and pass the full chat to a named human so the customer does not repeat themselves. A system with no human handover is not one Tarek builds.
In practice, the receptionist stops being the person who answers every first message and becomes the person who handles the exceptions, with the agent doing the sorting beforehand. Tarek Abou Rjeily, an AI consultant and builder in Beirut, designs AI agents for WhatsApp with human-approval steps where money or reputation is at stake.
Start small. Automate one path, such as after-hours first reply and booking, run it for a few weeks with staff reading the transcripts, and widen the scope only after you trust it. If your volume is a handful of chats a day, a receptionist or the owner alone may be all you need, and an agent is not yet worth building.
For repetitive first replies and bookings, it can take a lot of the load. For complaints, judgment and closing, it should not replace a person. Most businesses end up redefining the receptionist's role rather than removing it.
They should. Tell them it is an automated assistant and make it easy to ask for a person. Customers accept automation for simple tasks and resent being stuck without a way out.
Modern language models handle a lot of Arabic and mixed-language messages, but quality varies by dialect and by topic. Test it on a sample of your own real chats before you rely on it, and expect a staff review period.
Plan for it. Keep the agent to a fixed set of tasks, log every chat, review transcripts regularly, and let it hand over when unsure. A wrong answer on opening hours is minor, so the risky topics should go to staff.
Yes, more so. When customers write across time zones and weekend patterns, round-the-clock first replies matter, and a human team can then pick up the qualified chats during its own working hours.
Send the question on WhatsApp and I'll tell you honestly whether it needs building, buying, or fixing by hand.
Ask on WhatsApp →