Which tool to choose
The right question isn't "AI or programmer", but "how much does it cost if this thing breaks?".
- If a failure costs your time and nothing more (a script you use yourself, a personal automation, a draft), the AI is more than sufficient. You test it, if it doesn't work you fix it, nobody else suffers.
- If a failure costs money, data or reputation of clients or the company, you need someone who answers for what's running: a programmer who reviews it, secures it and takes responsibility for the code.
The dividing line is responsibility, not difficulty. The AI writes complex code; what it can't do is answer for it when, six months from now, it stops working in the worst way and at the worst moment.
How to decide
It isn't a gut choice: you decide by answering a few honest questions.
- Ask yourself who uses the thing you're building. Only you? Probably the AI is enough. Paying clients or colleagues who work on top of it? The bar goes up.
- Ask yourself what happens if it breaks silently. A wrong calculation in a spreadsheet you use yourself you correct; a wrong calculation in a quote sent to a hundred clients is serious damage.
- Ask yourself whether it touches sensitive data or payments. Personal data, credit cards, logins: here security isn't a detail the AI improvises, it's a craft.
- Ask yourself whether it has to last and grow. A throwaway trick you do with the AI; a system you'll use for years and that will have to handle more users deserves a solid base from the start.
- Use the AI itself for the first verdict. The operating syntax:
I want to build this: [describe what you want to do and who will use it].
Tell me honestly: can I do it myself with your help without knowing how
to code, or do I need a professional programmer?
Explain to me the concrete risks if I did it alone, in particular
on security, data and durability over time.
- Check: if the AI itself, describing the risks, names security, payments, other people's data or "many users at the same time", take it as the signal that an expert is needed. The AI is honest enough to tell you when it's beyond safe to rely on it alone.
A concrete example
Roberto has an idea: a site where the clients of his practice book appointments and pay the deposit online. He starts with the AI and in one afternoon he has a page that shows the calendar. It works, and he's thrilled.
Then he asks the AI to add the payment. The answer holds him back: handling real payments means touching credit card data, complying with security rules, preventing someone from booking without paying. The AI writes him that for the showcase part he can go on alone, but for real payments he needs a programmer to connect a certified payment system and secure the whole thing. Roberto keeps the part he did himself and entrusts only the money piece to a professional. He spends where it matters, saves where he can.
When it does NOT work (and how to fix it)
If you've built everything with the AI and now you don't know how to keep it standing
It's the risk of do-it-yourself pushed too far: you have a system that works but that nobody understands, not even you. Before it becomes a problem, have the code reviewed by a programmer on a short engagement: "look at what I built, tell me the risks and what needs fixing". Paying for a few hours of review costs less than a disaster.
If a programmer tells you it all has to be redone from scratch
It happens with projects born to grow but built like throwaway prototypes. It isn't always the professional padding the bill: beyond a certain size, an improvised base really does need rebuilding. Ask for a second opinion and have them explain what exactly doesn't hold up and why. If the reasons are concrete (security, durability under load), believe it.
If you don't have the budget for a programmer but the project is serious
Don't throw yourself into do-it-yourself anyway for the risky pieces. Reduce the ambition: start from the part you can safely do alone (the showcase, the information) and postpone the critical part (payments, client data) until you have the resources to do it well. A small, secure project beats a big, fragile one.
If the AI assures you it can do everything on its own
Beware of overconfidence, yours or its. The AI writes the code of a payment system, but "writing the code" and "guaranteeing it's secure against those who try to force it" are two different things. Ask it explicitly: "what are the security risks if I put this into production without an expert checking it?". The honest answer brings you back down to earth.
A tip from someone who actually uses it
Use the AI to quickly reach a prototype that works, then use that prototype to talk to a programmer. It's the cheapest way to work with a professional: instead of paying them to figure out what you want, you show them something concrete and they step in only where it really matters. The prototype made with the AI isn't the final product, it's the clearest quote request you can bring to an expert.
Frequently asked questions
Can a programmer start from the code I made with the AI?
Yes, and it's often a good starting point, provided it's readable. Keep everything you built and the prompts you used: they help the professional understand what you wanted. Sometimes they'll rebuild parts from scratch, but your work will have shown them the direction anyway, and that's paid for in fewer hours.
Will the AI make programmers disappear?
No, it changes the craft. The AI writes the code ever faster; the hard work stays human: understanding what's really needed, choosing how to build it so it holds up over the years, securing it, fixing the failures that don't resemble any manual. The AI has raised the bar of what you can do alone, it hasn't erased the zone where an expert is needed.
How do I know if a programmer is good if I don't understand the code?
Not from the code, but from the questions they ask. A serious professional, before writing a line, asks you who will use the thing, what happens if it goes down, how many users are expected, how you handle the data. Someone who starts writing without asking you anything is the warning sign. The right questions are worth more than a resume.
So what do I need the AI for, if for serious things an expert is needed anyway?
To flip the balance of power. Without AI, faced with a technical project you were completely dependent on whoever built it: you didn't know what was feasible, how complex it was, whether they were padding your bill. With the AI you arrive prepared: you have a prototype, you understand the terms, you can tell the easy piece from the risky one. It doesn't replace the programmer for serious things. It makes you a client who no longer gets taken for a ride.