And everyone promotes them for tasks they aren’t experts in.
Managers think they could replace devs, but never a manager.
Devs think they could replace management but never a senior developer.
Storyboard drawers think they can write screenplays. Screenplay writers think they can draw storyboards. Etc.
As an expert, you know how shit AI is in your own field, but surely those other jobs are simple enough to be replaced.
It’s very easy to replace something that was never critical to the process in the first place. My manager essentially updates my git tickets with what I did. We talk for 5 minutes a week. He just kinda lets me do my thing, I am fully aware of how lucky I am.
The principal task of a competent manager is, primarily, intervening between incompetent upper managers and actual workers. Replacing the incompetent manager removes the need for the competent one.
They’re incredibly useful, but you have to treat their output as disposable and untrustworthy. They’re reinforcement trained to generate a solution, regardless of if it’s right, because it’s impossible to AI evaluate that these solutions are correct at scale.
If you’re writing some core code: you can use an agent to review it, refactor parts, stump the original version, infill methods, and to run your test/benchmark scripts.
but you still have to manage it, edit it, make sure it’s not recreating the same code in 6 existing modules, generating faked tests, etc.
As an example this week on my side project I had Claude Opus write some benchmarks. Total throwaway code.
It actually took my input files, generated a static binary payload from it using numpy, and loaded that into my app’s memory (on its own that’s really cool), then it ran my one function and declared the whole system 100x faster than comparable libraries that parse the original data. Not a fair test at all, nor was it a useful test.
You cannot trust this software.
You’ll see these games metrics, gamed tests, duplicate parallel implementations, etc.
Hahahaha this idiot thinks that it’s the speed of our typing that ships code faster. He’s in a knowledge shortage.
I’ve found that the people who understand these “agents” the least are the ones who are promoting them the most.
And everyone promotes them for tasks they aren’t experts in.
Managers think they could replace devs, but never a manager.
Devs think they could replace management but never a senior developer.
Storyboard drawers think they can write screenplays. Screenplay writers think they can draw storyboards. Etc.
As an expert, you know how shit AI is in your own field, but surely those other jobs are simple enough to be replaced.
Let’s be honest, though: they absolutely could replace management.
It’s very easy to replace something that was never critical to the process in the first place. My manager essentially updates my git tickets with what I did. We talk for 5 minutes a week. He just kinda lets me do my thing, I am fully aware of how lucky I am.
some management, sure. just like some of your coworkers could probably be replaced by AI. but not the competent ones, and not the essential ones.
and personally, I’d still rather work with an incompetent person who can improve than four incompetent chatbots
although I’d rather work with no incompetence at all
The principal task of a competent manager is, primarily, intervening between incompetent upper managers and actual workers. Replacing the incompetent manager removes the need for the competent one.
90% of my experience with management is having none at all would be a net benefit
why would we want to add ai to that mix
This.
They’re incredibly useful, but you have to treat their output as disposable and untrustworthy. They’re reinforcement trained to generate a solution, regardless of if it’s right, because it’s impossible to AI evaluate that these solutions are correct at scale.
If you’re writing some core code: you can use an agent to review it, refactor parts, stump the original version, infill methods, and to run your test/benchmark scripts.
but you still have to manage it, edit it, make sure it’s not recreating the same code in 6 existing modules, generating faked tests, etc.
As an example this week on my side project I had Claude Opus write some benchmarks. Total throwaway code.
It actually took my input files, generated a static binary payload from it using numpy, and loaded that into my app’s memory (on its own that’s really cool), then it ran my one function and declared the whole system 100x faster than comparable libraries that parse the original data. Not a fair test at all, nor was it a useful test.
You cannot trust this software.
You’ll see these games metrics, gamed tests, duplicate parallel implementations, etc.
spend more time fixing slop compared to just doing it manually and correct the first time