A carpenter shows you a shelf.
Do you ask if they used an axe or a power saw?
You don’t.
You ask if the shelf holds weight.
Nobody has ever paid a carpenter for the swing of the axe.
They paid for a shelf that doesn’t collapse under a stack of books.
Manual coding errors are mostly gone now.
Typos, syntax slips, the small stuff that used to eat an afternoon.
But here’s what nobody’s saying out loud.
The mistakes didn’t disappear.
They moved.
An axe is slow.
You feel a bad swing before the log splits wrong.
A power saw doesn’t slow down for a bad cut.
It goes through the wood at full speed, mistake and all.
That’s the actual shift.
Not that AI writes code.
That AI writes ten shelves in the time you used to build one, and ships all ten before you’ve checked if a single one holds weight.
The old bug was a typo.
The new bug is a wrong assumption, built fast, built confidently, and built at a scale you can’t eyeball anymore.
So the job changes shape.
Not “review every line.”
Catch the crack before the shelf leaves the workshop.
That means guardrails before you start.
Tests that check the assumption, not just the output.
A habit of asking what this thing assumed about the world, before you ask if it runs.
The carpenter’s real skill was never the axe.
It was knowing which cut mattered and which shelf couldn’t afford to be wrong.
That part didn’t get automated.
It just got more expensive to skip.
So here’s my question for you.
Not a gotcha, an honest one.
When was the last time you checked the assumption instead of the syntax?
Leave a Reply