I thought the constraint was the software

For two years I optimised the part you can change in an afternoon.

The model, the pipeline, the way the data arrives: all real things, all things you can redo on Friday, if by Friday you have understood them better.

Then we sized the first plant on the client's consumption peak, and that was wrong.

The peak that counts is the grid's.

The fix wasn't a deploy: it was an order with an eighteen-month lead time.

What I hadn't understood is dull to say out loud: the constraint sits where time doesn't compress. Everything else is preference.

Software is the part of the system that forgives you. That's also why we spend so much time in it.

It isn't a general lesson and I wouldn't turn it into a principle.

It's just the reason why now, when I look at a new project, the first question I ask isn't about the stack.