The fastest way to vet an AI agent developer in Dubai is to ask for one thing: a production system handling real traffic, with numbers attached, that they can explain end to end. Everything else in this checklist expands on that test. I am writing this from the other side of the table — I am the kind of hire this post describes — so treat it as an insider's answer key: these are the questions I would want a founder to ask me, because they are the ones a ChatGPT-wrapper reseller cannot answer.
Why the title is confusing right now
'AI agent developer' currently covers at least three different people. A prompt engineer writes and refines instructions for models — a real skill, but only one layer of the job. A no-code automator wires tools like Zapier or n8n together — useful for integrations, but often without the engineering depth to handle failure states. An AI agent developer builds the whole system: orchestration, tool calls against real APIs, error handling, evaluation, and monitoring. When you post a role or brief a freelancer, all three will apply, and their day rates and outcomes differ enormously. The rest of this checklist is how you tell them apart in one conversation.
The six questions to ask before hiring anyone
One: 'Show me an agent you built that is in production right now.' Not a demo video — a live system with real users or real business traffic. Two: 'What happens when the model gets it wrong?' You are listening for fallbacks, retries, human escalation paths, and confidence thresholds; if the answer is 'the model is very accurate', end the call. Three: 'Who owns the code and prompts when we are done?' The right answer is you, in a repository you control. Four: 'How will we measure whether this worked?' A real developer proposes metrics before you ask. Five: 'What is the stack, and why?' Vague answers about 'proprietary AI' usually mean a thin wrapper on someone else's API. Six: 'What is a realistic timeline, and what has slipped for you before?' Anyone who has shipped real systems has a slippage story; anyone who promises two weeks for everything has not.
What a real portfolio shows — and the red flags
A credible portfolio contains live case studies with metrics: what was built, what it replaced, what changed, expressed in numbers the client would recognise. Red flags: screenshots of ChatGPT conversations presented as 'AI systems', portfolios that are entirely mockups or 'concepts', metrics with no context ('increased efficiency by 300%' of what?), and an inability to demo anything live on the call. Ask them to walk you through one project's failure handling specifically — the part demos never show — and watch whether the answer is architectural or hand-wavy.
Freelancer vs boutique agency vs in-house
ONE TACTIC A WEEK
For a UAE SME budget, the tradeoffs are mostly about accountability and continuity. A freelancer is cheapest and fastest to start, but you carry key-person risk: if they disappear, so does the knowledge — which is why the code-ownership question above matters most in this scenario. A boutique agency costs more but gives you process and some redundancy; verify the person on the sales call is the person doing the work. In-house is the most expensive and slowest to hire in this market, but right once AI systems become core to how the business operates rather than a project. Pricing across all three varies widely because the market is young and titles are unstandardised — which is exactly why you vet on evidence, not rate card.
A worked example — what a real answer looks like
When someone asks me 'show me your work', the answer that lands is Printo, a UAE printing company: 3,750 conversions on AED 42K of ad spend, with an agent stack behind the account doing outreach, quoting, and reporting work that previously ate human hours. The point is not the specific numbers — it is the shape of the answer: a named client, a real budget, a countable outcome, and a system I can open up and explain layer by layer, including where it failed early and what I changed. That shape is what you are hiring for. Any candidate worth engaging has at least one story with that shape; most pretenders have none.
Contract and ownership essentials
Put in writing: you own the code, the prompts, the workflow configurations, and the data pipelines at the end of the engagement, delivered in repositories and accounts you control. Prompts are the piece founders forget — they encode months of iteration and are trivially lost if they live only in the developer's tooling. Also settle API accounts up front: model and tool subscriptions should be on your billing, in your name, so nothing breaks when the engagement ends. A developer who resists any of this is planning to rent you your own system.
Onboarding — what access is legitimate
A real build needs real access, and a structured developer will ask for it in a structured way: API keys delivered through an environment file or secrets manager (never pasted into chat), ad account access at the appropriate permission tier, CRM access scoped to the objects the agent touches. What should stay gated: banking and payment credentials, owner-level admin on anything, and customer data beyond what the agent genuinely processes. My own client onboarding runs as a checklist for exactly this reason — access requests with a stated purpose for each item. If your candidate's onboarding is 'just give me your logins', that tells you everything about how the rest of the engagement will run.