Prompts are not code: if you understand none of the generated code, every argument about the program is a sophisticated version of “I told it to make no mistakes”.
0 Boosts
0 Likes
They're even slopping the requirements in one (large customer 😧) project I've until recently been assigned to.
And bragging about it.
So much for "this will help us concentrate on the big picture".
Gain in measured velocity, by the way, AFAICT: zero.
(At this point, I'm more flabbergasted than complaining! Such widespread complying-in-advance wasn't on my crystal ball.)
@quincy we did an experiment with letting someone slop an issue. We first estimated the time we’d need when doing it ourselves. To be fair we used the upper limit.
The expectation was that it would go faster.
It was slower.
EDIT: and horrifyingly that did not lead to people saying “we tried, it failed, let’s not do that anymore”.
(and same here about the complying ☹)
@ArneBab I wonder when this will start to register ... the delusion might be widespread, but could they please look at those darned KPI for *once*!
@quincy I think it’s because people always find a way to justify continued usage. It’s last month’ tech (that argument didn’t change for the past two years), it’s prompted wrong, the code isn’t clean enough, …
And only the arguments for continued usage get that leeway.
Did you see my quick re-evaluation of the experiment by metr? They claimed that they must have missed part of the speed gains, but did not take into account that repeat participants may have gotten worse:
https://rollenspiel.social/@ArneBab/117520574197499133
@ArneBab
My (soon to be be ex-, unless the bubble bursts this week *and* the machine idols go 404) coworkers no longer understand this.
They've fallen in lockstep behind that Claude thing.
The pipeline from "I'm just letting it make changes" to "we no longer want to understand the code or the process" was quick.
(This is why I'm so against even assigning this tech even the smallest role in a project. It's poison.)