We would still be coding in QBasic

I started with QBasic. Since then, every new tool has been billed as the end of software development: Microsoft Publisher, PowerBuilder, then low-code and no-code were each going to make developers unnecessary. None did. Each one took away a task, never the craft: we used to compile by hand, and today the tooling does it without our noticing; we used to set up every server by hand, and now infrastructure is written as code and reads like code. What used to be the job became a detail, and the job moved up a level. Had technology not evolved, we would still be coding in QBasic. Nobody misses it.

What we miss is never the tool, it is the mastery we had of it

So generative AI does not worry me: it reassures me, on two conditions. Lay the foundations first: a named domain, explicit invariants, tests that state what the system refuses, a review that does not rubber-stamp. Then guide the change: teams trained to ask, to read and to refuse, not left alone with a tool that says yes to everything. Without those two, you very quickly get applications that run, and that nobody masters.

So the question that matters is not whether AI will replace someone. It is how to channel its power into value, rather than into ever more code that works without anyone knowing why. QBasic did not teach me to type; it taught me to want to understand what the machine did with my lines. That is the one thing I have kept from one tool to the next.

AI is not a worry, it is a reassurance: provided you know what you ask of it