I Don't Need Another Fucking Job Managing AI
Automation was supposed to eliminate repetitive work. Somehow it created an entire department.
I love automation.
Always have.
If I find myself doing the same thing repeatedly, sooner or later I'll start wondering why the hell I'm still doing it manually.
That's one of the reasons I've spent so much of my life building software.
Computers are supposed to handle repetitive operations.
They're supposed to execute instructions.
They're supposed to make things easier.
And artificial intelligence promised something even more interesting.
Instead of explicitly programming every little detail, I could describe an objective and let the system figure out how to accomplish it.
That sounded wonderful.
I could spend more time thinking, designing, creating, experimenting, and doing the parts of engineering that I actually enjoy.
Less mechanical work.
Less repetitive bullshit.
More progress.
So I started building workflows around AI.
Skills.
Instructions.
Repository conventions.
Project documentation.
Automated processes.
Different tools for different purposes.
I wasn't simply chatting with a machine anymore.
I was constructing an environment in which these systems could operate with some degree of consistency.
And much of the time, it worked.
Sometimes impressively well.
That's important.
I'm not sitting here claiming AI can't do anything useful.
I've watched it accomplish complicated technical tasks that would otherwise consume substantial amounts of my time.
That's why I keep using it.
That's why I keep building around it.
But the problem becomes obvious when the system stops following the intended workflow.
Because then a whole new category of work appears.
First, you have to explain what you want.
Then you have to make sure the AI understood it.
Then you have to inspect the plan.
Then you have to watch the implementation.
Then you have to verify the output.
Then you have to check whether the implementation actually matches the original objective.
Then you discover that the AI introduced something you never requested.
Now you have to debug that.
Then you have to correct the instructions so it won't happen again.
Then you have to test the corrected instructions.
Then, the next day, it does something completely different.
And now you're doing all of it again.
Congratulations.
You've automated the original job and invented six additional responsibilities.
What makes this especially absurd is the sheer amount of infrastructure I already maintain to support the automation.
Repositories.
Skills.
Agents.
CI pipelines.
Documentation.
Tests.
Configuration.
I already had enough fucking engineering work before AI came along.
Now I have another layer of engineering dedicated to supervising the thing that's supposed to reduce my engineering workload.
And when something goes wrong, the assistant often provides an elaborate explanation of why it happened.
Except sometimes that explanation is wrong.
So now I have to debug the explanation too.
Wonderful.
We've achieved recursive bullshit.
The AI makes a mistake.
The AI explains the mistake.
The explanation contains another mistake.
I challenge the explanation.
The AI apologizes and produces a revised explanation.
Then I have to verify whether the revised explanation corresponds to anything that actually happened.
At some point, you have to wonder whether you're still doing software development or participating in an extremely elaborate experiment in administrative suffering.
And the cruelest part is what this does to motivation.
Because when I'm working on something interesting, I want to keep moving.
I want to build.
Experiment.
See results.
Enjoy the process.
I don't particularly care whether I'm the best programmer in the world.
I don't need some grand achievement waiting at the end.
I enjoy doing the work.
But nothing kills that enjoyment quite like spending an entire evening arguing with an automated assistant about the meaning of instructions you already wrote and tested.
Suddenly the project isn't moving.
The creative work isn't happening.
The automation isn't saving time.
And all the energy I had for doing something interesting has been consumed by managing the machine.
That's not the future I was looking forward to.
I don't need artificial intelligence to be perfect.
I don't need it to agree with me.
I don't even need it to solve every problem.
I need it to do the fucking job I gave it, tell me when it can't, and avoid inventing a completely different assignment along the way.
Because I already know how to manage unreliable systems.
I've spent twenty-five years doing exactly that.
What I wanted from AI was a way to spend less time managing them.
And now, after all this supposedly revolutionary progress, I find myself staring at the computer and thinking:
I don't need another fucking job managing AI.
I already have one.
It's called being a software engineer.
And apparently, according to the experts, artificial intelligence is going to replace me.
Excellent.
I'd love to see it.
Just as soon as I finish debugging the fucking replacement.