Buyers ask us this every week, and it is a fair question. AI tools really do write most of the code now. Here is exactly what that made cheaper, what it did not touch, and how our prices changed as a result.
A business owner in Sharjah put it to us plainly last month: "I read that AI writes the code now. Why is your quote still four figures?"
It is a fair question, and the premise is right. In August, Anthropic made the automatic mode of its Claude Code tool the default, meaning the AI takes most actions without asking. Its own figures show developers approve 97 percent of the permission prompts they see, and its automated safety classifier caught nearly 90 percent of dangerous commands in testing, where human reviewers caught 14 percent. Gartner expects 90 percent of enterprise engineers to be using AI assistants by 2028. Inside our own team, AI writes the majority of the lines in a typical project.
So what are you paying for?
Writing the code. Genuinely, dramatically. Screens, forms, database tables, integrations with well-documented services, tests: what took a developer a day takes an hour or two of supervised AI work. This is real and it is why our prices for standard projects fell over the last two years while the projects themselves got bigger.
The first draft of everything. Documentation, migration scripts, the fiddly setup of a new project. All faster.
Fixing bugs in familiar code. An AI that can read the whole codebase finds the cause of a bug faster than a person scrolling.
Understanding your business. The AI does not know that your Tuesday stock count works differently from the Friday one, or that "approved" means three different things in three departments. Someone has to sit with you, ask the questions, and write it down in a form that can be built. That is the discovery and scoping phase, and it is where projects succeed or fail. It takes the same time it always did.
Deciding what not to build. The most expensive software is the software you did not need. An AI will happily build every feature you mention. An experienced person tells you which three of your twelve requests will get 90 percent of the value, and that conversation is the difference between a $4,000 project and a $12,000 one.
Design. Not decoration; the decisions about what a screen shows, in what order, so that a warehouse worker on a phone gets it right the first time. AI produces plausible screens quickly. Producing the right ones still takes a designer who has watched people use them.
Responsibility. When something breaks on invoice day, someone has to answer the phone, understand what happened and fix it. The code being AI-written does not change who is accountable, and accountability costs money because it is a person's time on call.
Data migration. Getting five years of records out of a spreadsheet into a clean database is still mostly patience and judgement. The AI helps. It does not make the decisions about what the fourteen blank rows in 2023 mean.
Security and the guardrails around the AI itself. Ironically, the more AI in a project, the more review it needs. We wrote about the five rules we follow for AI agents that touch business systems. Applying them takes time.
| Project | Typical price, 2024 | Typical price, 2026 | What changed |
|---|---|---|---|
| Business website, 5 to 8 pages | $1,200 to $2,000 | from $500 | Build time fell by two thirds |
| Spreadsheet replaced with a multi-user app | $2,500 to $5,000 | $800 to $2,500 | Build time fell; scoping time did not |
| Custom ERP, 3 to 5 modules | $8,000 to $20,000 | $1,500 to $12,000 | Build fell; discovery, migration and training did not |
| Mobile app, single purpose | $3,000 to $6,000 | from $800 | Cross-platform code plus AI |
In AED multiply by 3.67, in GBP by 0.79. The pattern is the same in every row: the building got cheap, the thinking did not, and the thinking is now most of the price.
If a quote for a standard business app is still at 2024 prices, ask what the development phase costs on its own. If it is the majority of the total, the vendor has not adjusted. If a quote is suspiciously low, ask how much time is in discovery and testing. If the answer is "we start building straight away", you will pay the difference later in changes.
A fair quote in 2026 spends more of its budget on the conversation before the build and the support after it than on the build itself. That is what ours look like, and we send them as a fixed price within 48 hours of a call.
See the full breakdown in how much custom software costs in 2026, or book a free call and ask us the Sharjah question yourself.
We build the systems described in this article. Let’s talk about your project.
Keep reading