On AI Coding
You write a solid prompt.
You expertly engineer the context around the problem, about which you have some insight.
You direct the model's attention carefully and deliberately.
You plan, and replan, working in steps and answering questions you didn't initially have.
You build up a perfectly clear picture of what the outcome needs to be, iteratively, with your model of choice.
Sure, you're now using a tool that takes you some steps away from the statements, the loops, the functions.
And you're augmenting your brain, your ability, your knowledge, your hard-won intuitions as an engineer with a tool that gives all of those things more leverage.
But you're not outsourcing your knowledge, nor your understanding of the problem.
You're exploring it the same way you did before, but inside a new kind of tool that can execute faster for you in many cases.
How well it does that largely still depends on you.
AI coding isn't about thinking or understanding less. It's about leveraging that same thinking a lot more once the execution bottleneck is removed, and some of the details can be safely abstracted.
Just as they are when you write a function, and no longer consider where you need to move stack pointers, which registers are going to hold your return values, and what the endianness is.
Your understanding of the problem and the domain is what now, and always has, allowed you to navigate and solve for the right problems, decisions, tradeoffs... especially in the long tail of any hard thing.
You should never outsource that, unless you want a different, and much worse problem to solve.
However, if you stay in the driver's seat, using and building your understanding, independent thinking, and experience, now you have leverage you didn't have before.
Coding with AI is not just coding with AI... it's what you do between the lines that makes the real difference.