50 Years of Code wasn't wasted

18th October 2026

Thoughts from an old coder - where are we going?

From the BBC Micro to Claude Code

I wrote my first lines of code at twelve, on machines that most people under forty have only seen in museums. BBC Micros, Commodore PETs, computers with less memory than a modern email signature. I am sixty now, and there has not been a year in between without a keyboard in front of me.

Along the way I picked up around twenty programming languages, a senior career in engineering, a working fluency in mathematics, and a particular fondness for database design and automation. Schemas, frameworks, conventions, the unglamorous scaffolding that decides whether a system survives contact with real users. That is the part of the craft I have always cared about most.

Then, in the last couple of years, the ground moved. AI coding assistants arrived, and with them a question I keep being asked and keep asking myself: does fifty years of hands-on experience still count for anything?

What has actually changed

The short answer is that I no longer type most of my code. I describe what I want to Claude Code, and it wraps working software around my intent. The longer answer is more interesting, because what it wraps that software around is not just a prompt. It is my experience.

The productivity gain is real and, frankly, startling. User interfaces used to eat days of my life. Aligning components, chasing responsive layouts, tweaking the fiftieth form until it felt right. I simply do not do that anymore. I describe the screen, review the result, adjust in a sentence or two, and move on. The hours I have reclaimed from UI work alone would have been unbelievable to me five years ago.

But the speed is the least important part of the story. The quality of what comes out is where things get interesting, and that quality has very little to do with the tool on its own.

Why experience compounds rather than expires

An AI coder is extraordinarily capable, but it is also extraordinarily literal. Ask a vague question and you get a plausible answer. Ask a precise one, grounded in decades of knowing how systems fail, and you get something close to production grade on the first pass.

When I guide Claude Code, I am not just describing features. I am telling it how to normalise the schema, where the transaction boundaries belong, which naming conventions the whole codebase will follow, how errors propagate, and what the framework should look like so that the next person can read it. Those are not prompts I invented last week. They are lessons that cost me years, sometimes painfully, to learn.

The result is software that looks like it was written by a disciplined senior engineer, because in every sense that matters, it was. The AI supplies the typing. The judgement, the architecture, and the taste still come from the person holding the conversation. Experience has not been replaced; it has been amplified.

The gap that is coming

Here is what genuinely interests me, and slightly worries me. We are approaching a generation of builders who may never write code by hand at all. They will vibe code everything: describe, generate, run, repeat. Much of the time that will work beautifully, and I am not here to sneer at it.

But when it does not work, they will be entirely dependent on the same tools to debug, sense check, and fix the problem. The AI becomes an insanely clever black box, and nobody on the team can see inside it. They will not know what a well-formed schema looks like, so they cannot tell when they have been handed a bad one. They will not recognise a race condition, a leaking abstraction, or a framework that will collapse at ten times the load, because they have never had to live with the consequences of one.

Somewhere in the future there is a very real risk of a room full of clever people scratching their heads over a failing prompt, with no idea where the fault lies or how to even describe it.

Where that leaves people like me

I do not think the answer is to resist the tools. I use them every day and I would not go back. The answer, I suspect, is that deep experience becomes rarer and therefore more valuable, not less. Someone still has to know what good looks like, set the conventions, read the output with a critical eye, and step in when the black box shrugs.

That is a role, and it is a role that cannot be vibe coded into existence. It is earned the slow way, one broken system at a time.

So if I am still around when that generation hits its first truly baffling failure, I will be happy to help. My day rate, I expect, will be fabulous.