Comment by tibbar
1 month ago
On the other hand: sometimes doing the thing is itself a bad idea. One reason I continue to insist on design docs and code review is that I'd rather find this out ahead of time rather than deal with the damage afterwards.
In the GenAI era, "doing the thing badly without planning" has become so easy that some counterweight is needed.
The happy path is trivial now but I've found the gap between prototype and production is actually wider. You end up spending all your time handling non-determinism and latency issues that simply didn't exist with deterministic code. It seems like the real engineering challenge is just getting the unit economics to work.