AI agents in marketing: what should they do alone, and what must people approve?

AI agents have moved from suggesting ideas to doing the work. An agent can now research a topic, write a draft, check it against brand rules, schedule it and report on how it performed. That raises a practical question marketing leaders in the Gulf are asking: where should the agent stop and a person start?

What is an AI agent in marketing?

An AI agent is software that can carry out a multi-step task towards a goal, using tools and checking its own work, rather than answering a single prompt. In marketing, agents typically research, draft, schedule, monitor and report. Several agents working together, each with one role, form an agent team.

Why does the line between agent and human matter?

Because marketing is public. A wrong number in an internal note is a mistake. A wrong number in a bank's Instagram post can be a complaint, a regulatory issue and a news story. The more public and permanent an action, the more it needs a person.

A simple test works well: the more public, financial, personal or irreversible an action is, the more human approval it needs.

Which marketing tasks can AI agents do alone?

These tasks are internal, reversible and easy to check:

• Research and monitoring: audiences, competitors, trends, keywords, news and mentions.
• First drafts: posts, captions, scripts, articles, press releases and reports. A draft is not a decision.
• Formatting and versions: resizing, subtitles and platform-specific versions of approved work.
• Scheduling approved content: once a person has approved it.
• Routine replies: answering common questions using wording approved in advance.
• Reporting: pulling data and summarising it in plain language.

Which tasks need a senior marketing person?

These involve judgement, taste or relationships:

• Every piece of content before the client sees it. A person checks tone, facts and fit.
• Every media pitch and press release. Journalists deal with people, and should.
• Sensitive or off-script replies: complaints, health, money, politics, religion and anything unusual.
• Strategy changes: new positioning, new audiences, new channels.

Which decisions belong only to the brand owner?

• Anything published in the brand's or a person's name.
• Any spend: ad budgets, boosts, creator fees.
• Any use of a real person's face, voice or likeness, which also needs that person's written consent.
• Crisis responses. In a crisis, no automated reply should go out.

What should never be automated?

• Publishing content no one has approved.
• Fake reviews, bought followers or fake engagement.
• Creating a synthetic version of a real person without their consent.
• Sharing one client's data or materials with another.
• Agents discussing a customer's personal account or medical details.

How do regulated sectors adapt this framework?

Regulated sectors add approvers rather than change the framework.

• Banking and finance: compliance becomes a named approver, and agents draft from the product material the bank approves.
• Government: approval chains mirror the entity's hierarchy, and Arabic content is written natively for Arabic readers, not translated, and a senior Marca person approves it.
• Healthcare: medical reviewers approve every claim, and patient questions always go to a person. Health advertising also needs a licence from the Ministry of Health and Prevention (MOHAP) before it runs; Abu Dhabi facilities log their health ads with the Department of Health (DoH); and in Dubai, the facility's Medical Director must approve social-media content.
• Influencer work in the UAE: individuals who publish advertising content on social media or other digital platforms need an Advertiser Permit from the National Media Authority (formerly the UAE Media Council), and the brand should check it before any content goes out. People promoting their own business on their own account are exempt. Visiting creators can get a permit only through an agency accredited by the National Media Authority, valid for 3 months and extendable to 6. Ads for health, real-estate, financial or education services also need approval from the sector regulator.

How does an agency prove its agents are under control?

Ask any agency, AI-run or not, five questions:

1. Can you show me, in writing, what your agents do alone?
2. Who approves content before I see it, and how senior are they?
3. How is AI use disclosed to us and to the public?
4. Who owns the brief and guidelines the agents work from?
5. What is your policy on synthetic faces and voices?

If the answers are vague, the control probably is too.

How does Marca apply this?

Marca publishes this framework as its Autonomy Charter on marca.ae/trust. AI agent teams do the research, drafting, scheduling, routine moderation and reporting. A senior Marca person reviews every output. The client approves everything published, every spend and any use of a likeness. Routine replies follow a library approved in advance by a senior person; everything else needs a human yes.