A 9.9 lands on a Tuesday
One Fiori role could run whatever ABAP it felt like on a production ERP. Ten years ago the expensive part of that day was working out whether it applied to you. That's a report now. What's left is the signature, and what's left keeps getting smaller.
Second Tuesday of the month, which anyone who has been in this ecosystem a while reads as a small knot in the stomach.
October delivered one that earned it. Someone holding one particular Fiori role could run whatever ABAP they felt like on a production ERP. Scored 9.9. Nobody asks whether that one can wait for the next sprint.
So you have hours, not weeks.
Ten years ago the expensive part of a day like that was working out whether the thing even applied to you. Which systems, which support package levels, which components were genuinely installed rather than just drawn on somebody's architecture slide. That took a morning, and it took a person who knew the landscape well enough that the answer could be trusted.
That's a report now. The ticket arrives already knowing which systems are in scope.
Nobody called that automation at the time. There was no announcement, no transformation programme, no slide. The work simply stopped existing, and the people who used to do it moved on to the next thing without quite noticing they'd been relieved of something they used to put on a CV.
I think about that whenever somebody asks whether agents will eventually take over this kind of work. The question always comes in the future tense, and the first wave already came through years ago. We can just look at what it took.
It took the analysis. What it left was the sentence at the end of the day: production comes down at two in the morning on day three of the close, we accept the risk to the business, and my name is on the authorisation.
An agent can tell you the note applies. It can't get fired for when you apply it.
The uncomfortable part
Except that what's left keeps getting smaller, and I'd rather say that out loud than pretend otherwise.
A lot of what I used to call engineering judgement turned out to be pattern recognition wearing a senior badge. Reading a dump and knowing roughly where to look. Sensing that a note would break a custom exit. Knowing which vendor estimate to double. It felt like intuition at the time. Most of it was interpolation over a large sample, and interpolation over a large sample is the thing these systems are now better at than I am.
The applicability matrix felt like expertise too, right up until a report started producing it.
So I don't think the honest version is that humans hold judgement and machines hold labour. The boundary moves in one direction only, and a fair amount of what sits on my side of it today will turn out to have been pattern matching the whole time.
People in this industry use "judgement can't be automated" as a warm blanket. It isn't a blanket, it's a shrinking perimeter, and the only useful thing to do is know exactly where you're standing inside it.
So think about the last hard call you made at work, the one you were quietly proud of.
Was that judgement, or pattern recognition that hasn't been automated yet?
Pablo