# Vibe coding vs a development agency in Dubai: price, speed and what you give up
A Dubai development agency will usually quote from AED 40,000 for a first working version and ask for six to eight weeks. A fixed-scope AI-assisted build lands at AED 9,000 for an internal tool and AED 18,000 for an MVP, delivered in about two weeks. Both numbers are honest. They buy different products, and the cheaper one shifts several risks onto your side of the table.
I build and then maintain what we ship, including the systems our own business in Dubai runs on. What follows is the mechanics of both routes, including the cases where I would send you to an agency to pay the higher number.
Two different products share one word
The word "development" gets used for two things that have almost nothing in common commercially.
The first is a service contract. You hire a company that assigns a project manager, a designer, developers, a QA engineer, and sometimes a business analyst. You buy their process: requirements captured in writing, an architecture that anticipates growth, test plans, documentation, a warranty period, and a support agreement afterwards. The code is one deliverable among several.
The second is a build, and this is what vibe coding sells commercially. You describe a specific thing, a price is fixed, and two weeks later the thing exists and runs. There is no analyst, no separate QA department, no 60-page specification document. One engineer drives an AI model to produce most of the code and reviews every line of it before it touches data or customers.
Vibe coding, in a commercial context, means an engineer directs an AI model to write the bulk of an application while a human reviews the output line by line before deployment. In Dubai it typically ships an internal tool or prototype from AED 9,000 in around two weeks, and an MVP from AED 18,000, at a scope agreed and frozen in writing before work starts.
Comparing the two on price alone produces a wrong answer roughly half the time. The useful comparison is what each contains and what happens in month four.
What actually sits inside a Dubai agency invoice
Local development rates in the UAE sit around AED 400 to 600 an hour for a Dubai-based team, and roughly AED 90 to 180 an hour for offshore delivery with a local account manager. Those are reported market bands rather than anyone's published rate card, but the arithmetic they produce is worth doing.
An AED 40,000 quote at AED 450 an hour is under 90 hours of senior time. That is barely two working weeks for one person. So when an agency promises a functioning MVP in six to eight weeks for AED 40,000 to 90,000, either the delivery is happening offshore at a fraction of the local rate, or the scope is much smaller than the conversation implied. Both are legitimate. Neither is usually said out loud.
Here is what the invoice is actually paying for, and what happens when a line is missing.
| Line item | What it buys | Cost of skipping it | |---|---|---| | Discovery and requirements | Senior time turning your idea into written screens and rules | A scope argument in week four | | Architecture and data model | A structure that survives ten times the current data | A rewrite when volume grows | | Interface design | Screens designed before they are coded | Developer-drawn layouts you dislike | | QA | Test plan, device matrix, regression passes | You become the tester | | Project management | Tracking, meetings, change control, usually 10 to 20% of the build | Every coordination task lands on you | | Documentation | Handover pack, API notes, runbook | The next developer starts by reading code | | Security review | Auth, access control, secrets handling, dependency checks | A customer finds the hole first | | Warranty and support | A defined window where fixes are free, then a retainer | You chase a developer who moved on |
Post-launch support in this market typically runs 15 to 25% of the build cost per year, plus hosting. On an AED 90,000 build that is AED 13,500 to 22,500 annually before you add a single feature. Any quote that stops at the build number is describing one line of a three-year cost, which is the same trap covered in more depth in the piece on how app development companies in Dubai structure their quotes.
None of this is padding. If your project genuinely needs those eight lines, an agency delivering all eight for AED 90,000 is fair value. The question is whether your project needs them yet.
What a fixed-scope vibe coding build contains
Our version is deliberately narrow. Before anything is written, the scope becomes a list: every screen, every action a user can take, every integration, every report. That list is the contract. Price is fixed against it. An internal tool or prototype starts at AED 9,000, an MVP at AED 18,000, and delivery is around two weeks. Handover includes the repository, the deployment, and two weeks of fixes afterwards.
The part that matters technically: a live engineer reviews every line the model produces before it touches data or customers. That review is the whole reason the format is safe enough to sell. AI models produce code that runs and looks convincing while carrying an access-control flaw that no test will catch, because the test only checks that the correct user succeeds, not that the wrong one fails.
We do not include a business analyst, a design phase with three concept rounds, a QA department, or a 12-month roadmap. Those get removed on purpose, and their absence is most of the price difference. The full breakdown of what each tier covers sits in the guide on what a working app costs in Dubai, so I will not repeat the tables here.
If your requirements are still verbal and shifting, this format will hurt you. Write them down first, or buy discovery from someone.
Where the fast route genuinely wins
Internal tools are the clearest case. When the users are your own staff, the blast radius of a bug is an annoyed employee rather than a public incident. A booking tracker, a commission calculator, a quote generator, an approvals queue. These are the projects that sit in an agency backlog for months because the budget never justifies a full engagement.
Validation is the second. Spending AED 18,000 to learn whether anyone will use the thing beats spending AED 90,000 to learn the same. If the answer comes back positive, you hand an agency real usage data and their discovery phase gets shorter for it.
Then there is the spreadsheet that has outgrown itself. The requirements already exist inside the file and nobody needs to discover them. This is the highest hit rate we see, and it overlaps with the decision covered in custom software versus off-the-shelf tools.
Deadlines make up the rest. A tender response tool needed before the tender closes. A seasonal booking flow before the season starts. Six weeks is sometimes the same as never.
Pricing tables also miss a compounding effect. Software live in two weeks starts producing data in two weeks. By the time an eight-week engagement launches, the fast build has six weeks of usage telling you which features nobody touched.
We run this on ourselves. SkyLight's studio booking runs on a WhatsApp agent we built in-house that handles enquiries, quotes, booking, invoices and payment links, and writes into the CRM without a person in the loop. It is in production, it takes real money, and we maintain it ourselves. That is the honest version of a portfolio: the thing we sell is the thing our own revenue depends on.
What you give up
This is the section most vendors skip, so here it is straight.
The requirement freeze
The fixed price exists because the scope is fixed. Changing your mind in week two means a new scope and a new price. An agency with a project manager absorbs a certain amount of churn inside the engagement, and you pay for that flexibility in the retainer whether you use it or not.
Architecture depth
Two weeks buys a design that is correct for your current and near-future load. It does not buy multi-tenant billing, horizontal scaling, or a data model shaped around a product you have not defined yet. If you know you are building a platform rather than a tool, say so at the start and expect a different number.
Documentation volume
You get a README, an environment variable list, a data model note, and a deployment walkthrough. You do not get a formal specification, an architecture decision log, or a test matrix that an auditor would accept.
Security of generated code
Veracode's 2025 GenAI Code Security Report found that around 45% of AI-generated code samples introduced a weakness from the OWASP Top 10. That is why the human review step is non-negotiable, and why you should ask any vendor selling AI-assisted delivery a direct question: who reviews the generated code, and what do they check? The answer should name authentication, access control on every endpoint, secrets handling, input validation and dependency versions. Vague answers mean the review is not happening.
Test coverage
Critical paths get tested. Everything else gets a manual pass. On an internal tool that is a reasonable trade. On a payment flow serving the public it is a bad one, and the scope should say so before anyone starts.
The support question
A fixed-scope build ends. After the two-week fix window, somebody has to own the thing. Decide who before you sign, rather than in month three when something breaks on a Friday.
One practical addition for this market: Arabic and right-to-left support adds roughly 15 to 25% to front-end effort. Put it in the original scope if you need it. Retrofitting RTL into a finished interface is one of the more expensive changes you can request.
When the agency is the right answer
There are projects where the higher price produces the cheaper outcome.
Regulated data is the obvious one. Health records, financial services, anything where a formal QA trail and documented security testing are contractual obligations rather than good practice. You are buying the paperwork as much as the software, and the paperwork takes real hours.
Integration with legacy systems that carry no documented API belongs in the same bracket. Discovery alone can burn two or three weeks, and reverse-engineering a fifteen-year-old ERP does not compress under AI assistance, because the bottleneck is access and permissions rather than typing speed.
A twelve-month roadmap with a growing team also argues for the agency. If you already know there will be six releases and three developers by next summer, buy the architecture and the documentation now. Retrofitting structure later costs more than building it once.
Unformed requirements point the same way. If you cannot list the screens, you are not ready for a fixed price. Buy discovery first, then decide who builds.
Procurement rules close the list. Some UAE enterprise and government buyers require a supplier with a certain headcount, professional indemnity insurance, and a support SLA carrying penalties. No two-week build satisfies that.
If two or more of these apply to you, the AED 40,000-plus route will cost less over three years than a cheap build you end up replacing. I would rather say that plainly than sell you the wrong shape.
Handover: who owns the code, the repository and the accounts
This decides more about your long-term cost than the build price does, and clients ask about it last. The repository should be created inside your organisation account from day one, not moved over at the end. Same for hosting, the database, the domain, the payment gateway account and any model API keys. If those live in the vendor's accounts, you are renting your own product, and you will learn the terms of that lease at the worst possible moment.
Deployment has to be documented well enough that a different engineer can ship a change without calling the original author. Onboarding a developer into an unfamiliar codebase takes weeks, and most of that time goes into working out how the thing gets built and deployed. A readable data model, a working local setup and tests on the critical path cut that to days.
A few contract lines are worth insisting on regardless of who builds: who owns the intellectual property, what happens after handover, what a change request costs once the project is closed, and how long the vendor keeps access to your accounts.
Our own shape after handover is plain. Two weeks of fixes are included, then you either maintain it yourself, which happens often with internal tools, or keep a monthly line for changes. If the thing turns into an operational process rather than a piece of software, the automation format covers it at AED 6,000 setup and from AED 1,200 a month.
A short checklist for deciding
The two routes side by side, using reported Dubai market bands for the agency column and our published prices for the other.
| | Dubai development agency | Fixed-scope vibe coding build | |---|---|---| | First working version | AED 40,000 and up | AED 9,000 internal tool, AED 18,000 MVP | | Timeline | 6 to 8 weeks at the fast end, 2 to 5 months for mid-market scope | About 2 weeks | | Scope changes | Handled through change control, billed | Scope frozen, changes are a new quote | | Who works on it | PM, designer, developers, QA | One engineer plus AI, reviewed by hand | | Documentation | Full handover pack | README, data model, deployment notes | | After delivery | Retainer at 15 to 25% of build per year, plus hosting | 2 weeks of fixes, then a decision you make | | Right when | Long roadmap, regulated data, legacy systems, formal procurement | Narrow problem, internal users, deadline that will not move |
Run these questions before you request a single quote.
Who uses it? If the answer is your own staff, start with the fast build. If it is thousands of paying customers handling money on day one, buy the deeper process.
What breaks if this is wrong for two weeks? If the answer is "some inconvenience", speed wins. If it is "a regulator asks questions", it does not.
Can you already list the screens and the actions? If yes, a fixed price is achievable. If no, pay for discovery before you pay for code.
Is regulated or sensitive personal data involved? That pushes toward formal process and documented testing.
Who maintains it in month four? Name the person or the company. In my experience an unanswered maintenance question predicts abandoned software more reliably than any technical factor.
Have you priced year one rather than build day? Add hosting, changes, and support to whichever route you pick before comparing numbers.
If you want the questions answered against your actual situation rather than in the abstract, a Growth Audit from AED 3,000 covers the process, the systems you already run and where a build would actually pay for itself. If you already know what you want built and can describe it in a list, the scope and price for a fixed-scope build can be agreed in a single conversation.
