“AI did that” is not acceptable
Like most engineers, I started small. A couple of years ago I was using Copilot to scaffold code and bootstrap new projects, the boring setup that eats an afternoon for no good reason. Then it started fixing bugs and suggesting solutions, and the productivity gains were obvious almost straight away. I’m a tester and an engineer at heart, but my overriding goal has always been to get things done, so I leaned in. Today, AI writes about ninety-five per cent of the code I produce.
That number tends to stop people, so I’ll be clear about what it means.
The work moved to the command line
The number matters less than the change behind it. Over the same period I’ve moved off the editor and into the command line, mostly using Claude Code, and I bring the data, the sources and the systems to me rather than hopping between them. I connect the tools I need and work in one place, from a single environment.
“Vibe coding” isn’t the right description, because it’s more planned than that. “Conversational coding” is closer: I direct the work, I make the decisions, and the machine carries them out. Whatever you call it, this isn’t a new tool bolted onto the old job. It’s a different way of working, and it has changed what my day actually looks like.
Whoever ships it owns it
This is the part I see people get wrong. I own everything I produce. How I produced it is irrelevant. I might have used AI, I might not have, but the quality of that output, and the data inside it, is on me.
The phrase that worries me most is “oh, Claude did that”. It’s a red flag every time. The moment you treat the tool as somewhere to put the blame, you’ve stopped being accountable for your own work, and that is a real risk rather than a figure of speech. Nobody reviewing a system, a bug or a data breach cares which tool generated the code. They care who stood behind it.
What about engineers just starting out
I feel in control of all this, and I’m confident I understand everything that comes out of it. But I should be, given the years I’ve spent in the industry. My judgement and experience are worth more now than they have ever been, precisely because someone has to know whether the generated code is any good.
That is what I worry about for people starting today. How do you build the instinct to debate, push back on, correct and validate AI-written code if you have never had to write much of it yourself? Maybe it resolves on its own, and engineers who work this way from day one develop a different but equally sharp judgement. Or maybe we end up with teams shipping code that nobody in the room actually understands. And then the obvious question: what happens when the AI is down, or the price jumps to a level you cannot justify?
The answers are not new. Clear onboarding paths, real documentation, architectural diagrams that reflect what is actually there. We have always known these matter. They matter more now, not less.
So
I love working this way, and the productivity gains are real and larger than I expected. But none of it holds up without personal accountability. Use whatever tools make you faster, work however suits you, but own the output as if you typed every character yourself. Because as far as anyone else is concerned, you did.