Which is exactly what we’re seeing now with LLMs writing code.
Interestingly you’ll see the same building a mega bridge or skyscraper. As long as the plan has been done well, it can be assembled by relatively very unskilled labor.
Note the test plan and acceptance criteria could be one and the same, depending.
That "if" is doing an unrealistic amount of lifting, here, in practice, and that's probably why human coders have had a better run than monkeys.
Also I will observe that something being rare is not in and of itself an indicator of whether or not it is reasonable. Many things are done a certain way because that's the conventional approach or it's more cost effective or due to some logistical concern.
In the case of software development it was previously cheaper not to exhaustively specify everything up front but I'd argue LLMs have upended that longstanding status quo within only the past year.
We had humans that were smarter than monkeys, dealing with imperfect requirements. Now we have LLMs that are smarter than monkeys, dealing with imperfect requirements.
LLMs don't work because of good requirements input, they work despite the poor requirements they're given.
Nice sentence. But it has 2 problems:
1) Define "high enough" - There is where good requirements start, make it measurable!
2) Related to "you’ll see the same building a mega bridge or skyscraper". I've seen enough cases where "unskilled labor" messed up building something. You will reply "that is why I said 'Acceptance Criteria'!". My take is: you are making it far more complicated than it should, taking totally unnecessary risks, and it will be invariably more expensive in the long run.
I mean, I've seen enough examples of cases where the acceptance criteria was not properly checked, because also the ones doing the test were "unskilled" and had received a "perfect plan". (Just look at any infrastructure disaster we have every year)
And even in the best case: you discover the poured cement does not meet the specifications. What now? Blow the thing and start over? Better to have skilled labor.
Again, my 2ct. experience: hire experienced people, who actually have done the work all the way down to the lowest level.
"If you think knowledge is expensive, try with ignorance"
That being said,
> A software Engineer needs to have all phases of the SDLC in mind, and be thinking about and planning all of them. Code is a tiny piece of the puzzle.
I would not AT ALL call it tiny. Yes, is one part of MANY. But is your final delivery. Would be like a Toyota saying: “hey they don’t really have to understand what makes a good car, the cars we sell is a tiny portion in the whole development cycle”