I have been programming for about 45 years. I started on a system to detect burst pins in a fast breeder reactor, and I have written the software my businesses run on ever since. Today most of the code I put live is written by an AI. So when people ask whether AI will replace programmers, I am not guessing from the outside. My answer is no, but the job is changing faster than at any time I can remember, and not everyone in it will come through the change in the same place.
How I used to write software
Before 1998, our business ran on a CRM I wrote myself in a 4GL on SCO Unix, used from Wyse terminals. In 1998 we gave up on a bought package after three consultants had spent three months tailoring it with no end in sight, and I wrote our own in Visual Basic 6 and SQL Server instead. We had a working prototype in a month and the full system three months later. I wrote every line of it myself.
That was how it worked for decades. If you wanted software, someone sat down and typed it, one line after another, and the typing took most of the time.
How I write it now
In 2026 I rebuilt the same CRM for the fourth time, in PHP using the Laravel framework. It has grown to 122 tables and 61 screens. I built most of it with Claude Code, Anthropic's AI coding tool, which I run from the command line, and I also use Codex, OpenAI's equivalent. They do the junior work. The design and the decisions stay with me.
The difference in speed is huge. This month we found that two thirds of the searches on our websites found nothing. We worked out why, built a fix, tested it against every search from the past month and put it live, all in a day. A few years ago that would have been a project.
So yes, an AI now does most of the typing. If programming meant typing, it would already be disappearing.
What the AI does not do
The AI wrote most of the code, but the things that decided whether our CRM worked were mostly things it did not do.
Knowing what to build. Before writing any code, I designed the system on paper in October 2025 and went through the requirements in depth with the office team. Of the 58 changes we have since made to the database structure, only one was a genuine design mistake, and it was fixed within a day. That came from people telling me how they work, not from a model.
Knowing what people will actually do. The biggest complaint after launch had nothing to do with features. People forgot to press save, moved to another screen and lost their changes. I did not see it coming, and nor did the AI. The office told me, and I moved the screens to save every change the moment it is made, with an undo button. It made more difference than anything else I built.
Taking responsibility. In October 2026 a security problem was found in PHP, the language our CRM is written in. I took the system offline until the fix was out, because our tenants' data mattered more than a perfect uptime record. An AI can write code, but it takes no responsibility for it. Code that looks finished can still have holes in it, and somebody has to answer for them.
Noticing when something is quietly wrong. When we added AI to our website search, a few searches came back empty for no obvious reason. The logs showed the model was thinking out loud about what the visitor meant and running out of room before it gave an answer. It had also started "correcting" cheque to check. Neither problem announced itself. Someone had to look.
Putting it all together. Getting the server, Claude Code and the web server working together, so that 300 websites are served securely and renew their own certificates, was hard. The AI helped with every piece, but making the pieces fit, and keeping them running, is still a person's job.
Checking the AI's work
For complicated code we never take the first version from one AI. Claude Code writes it, OpenAI's model reviews it, and Claude fixes what the review finds. A second model from a different company is good at spotting what the first one missed, such as missing checks, odd inputs and things that will break later. I explain the method in Using Two AIs to Check Each Other.
Even then, two models can agree on something that is wrong. Anything that matters still gets read by someone who understands it. That is the part of programming that has become more important, not less: reading code critically, and knowing what a correct answer looks like.
Your situation may be slightly different. ask a question below ↓ and our editorial team will reply with our advice.
Who should be worried
I think the people most exposed are those whose whole job is turning a clear specification into code. That is exactly the work AI now does well, and does in minutes.
The part that worries me most is the juniors. The routine work that AI handles best is the work that used to train new programmers. Every experienced developer started by writing the dull parts and learning from their mistakes. If that work goes, the industry will have to find another way to grow the people who check the AI's work in ten years' time.
The people best placed are those who understand a business and its problems. They can now direct a very fast assistant, and one of them can build what used to need a team.
What it means for a small business
For a small business this is mostly good news. Building your own system used to mean hiring developers. Now one person who knows the business and can read code can build a lot, for very little. Our CRM runs on a server costing about £50 a month. I explain why I built it in Building Our Own CRM in Laravel and PHP.
The catch has not changed. Someone has to look after the system for years: fix it, update it, keep it secure and check the backups. A bespoke system with nobody who understands it is a liability, however it was written. AI makes software cheaper to write. It does not make it cheaper to own.
My answer
I do not think programming is disappearing. Typing code by hand is becoming a much smaller part of it. What is left is the part that was always the hard bit: understanding the problem, designing the solution, judging whether the code is right and taking responsibility when it is not.
If I were starting out now, I would still learn to program, because you cannot judge code you cannot read. But I would spend as much time learning how businesses work, how to design a system on paper, and how to keep one secure. Those are the skills an AI cannot take responsibility for, and they are the ones that will be in demand.
Ask Technology for Small Business a question
Ask our editorial team a question and we will reply with our advice. Tell us as much about your situation as you can: the more detail you give, the more useful our answer can be.
You do not need to use your real name. Please do not include your full address, phone number, email address, or the names of other people. We may edit or remove identifying details for privacy and legal reasons.
Comments are moderated before publication.