Code used to be the hard part of software (was it?). Now it isn't. The hard part is building systems that don't break (or rather break gracefully). With AI, writing code has become a byproduct of the real work.
If you use AI to build things, your job changes. You aren't typing (at least, not typing boilerplate); you are building the fences, the context, and the checks to see if the thing actually works.
The Strategy of Alignment
Success happens before anyone writes a single line of code. It comes from planning. You have to treat the AI like a smart but forgetful lead engineer. Talk through the decisions. Explain the tradeoffs. If you don't map out the core ideas clearly, you will only end up with garbage generated at high speed (and debugging AI garbage is much harder than debugging your own).
The Source of Truth: agents.md
Context is the only thing that matters (and the hardest thing to maintain). I keep a file called agents.md in every project. It is not a todo list. It is a shared brain. It keeps track of progress and instructions so the AI knows what the ultimate goal is.
Engineering is still the hard work. The code itself is merely the output (and usually the easiest part now).