It is a question I hear more and more from founders and small business owners:
"Why should I pay a developer when AI can write the code?"
It is a fair question. I use AI coding tools every day, and they are astonishing. But the question is built on a wrong picture of what a programmer is, and that wrong picture is expensive.
What the data actually says
Two facts that rarely appear in the same sentence:
- 84% of developers now use or plan to use AI tools, yet more of them distrust the accuracy of AI output (46%) than trust it (33%) (Stack Overflow Developer Survey 2025, summary).
- In a randomized trial with experienced open-source developers working on their own large projects, allowing AI tools made them 19% slower, while they believed it made them about 20% faster (METR, 2025).
The second result surprised everyone, including the researchers, and it comes with honest caveats: early-2025 tools, mature codebases, experts who already knew the code. On small projects and for beginners the picture is different. But it tells us something important: writing code was never the slowest part. Understanding what to build, and checking that what was built is right, is.
The programmer was never a typist
When I built Rafu, an offline stock and credit book for small shops, the hardest decisions had nothing to do with typing:
- Should the app ask for permission to read the shop owner's SMS messages to match mobile-money payments? It would be easier. It would also be a privacy disaster. We chose a share-and-paste flow instead.
- Money as decimals or as whole minor units? One choice creates rounding bugs that only appear at the end of the month.
- What happens when two phones in the same shop sell the last bag of sugar at the same moment, with no internet?
An AI will happily write code for any of these. It will not tell you that you asked the wrong question. It answers with the same confidence whether it is right or wrong, which is exactly why I wrote, in Arabic, that the most important skill with AI is checking it when it is wrong with confidence.
So who is the smart programmer today?
Not the fastest typist. Not the person who refuses AI out of pride. Not the person who pastes whatever the model returns.
The smart programmer today:
- Owns the problem. Talks to the shop owner, the student, the nurse. Knows which mistake would hurt a real person.
- Designs before generating. Decides the data model, the boundaries, the failure cases, then lets AI fill in the walls.
- Reads more than they write. Reviews AI output like a senior reviewing a junior: line by line, with tests.
- Knows when to say no. To a feature, to a permission, to a dependency, to a shortcut.
- Stays accountable. When it breaks at 2am, someone has to understand it. A model is not on call.
Then, AI or programmer?
My honest answer for anyone with a real project:
- Prototype or demo, to test an idea? AI tools alone can take you surprisingly far. Do it, and learn.
- Something people will trust with their money, health, data or livelihood? You need a programmer who uses AI well. Not instead of AI, and not instead of a programmer.
The best code in this era is not written by AI, and it is not written by humans alone. It is written by a person who knows exactly what they want, uses AI as a tireless assistant, and refuses to ship anything they do not understand.
I explored this in Arabic too: AI and your job, how to prepare instead of panic, and how to ask AI a good question.
Your turn
- If you hire developers: has AI changed what you look for in a candidate?
- If you are a developer: where has AI saved you real time, and where has it quietly cost you time?
- If you are learning to code: should beginners avoid AI for the first months, or learn with it from day one?
Saifuddin Ahmed builds Ishraqa Daily and Rafu, and founded OtwoX Digital Solutions. More at saifuddin.dev and LinkedIn.
I drafted this piece with an AI writing assistant, which felt like the right way to test its own argument. The decisions, the examples and the opinions are mine.

