I use AI every day. When it works, it feels like having a super fast intern who never sleeps.
When it does not, it feels like explaining the same thing to a very confident goldfish.
AI absolutely helps me move faster, especially on repetitive work. Boilerplate, refactors, first drafts, scaffolding, all of that is easier now.
But lately I have also been feeling the other side of it. The part people do not talk about enough. The mental load.
Why it feels heavy
1) The communication overhead is real
Even when the idea is clear in my head, I still have to translate it into the right prompt.
Then I read the output, notice where it misunderstood me, and try again.
Then again.
Sometimes I get what I want in one shot. A lot of times I do not. It is still faster than starting from zero, but it can feel weirdly draining because the work becomes more about steering than building.
2) AI only works well when the setup is good
AI gets much better when I give it clean requirements, constraints, examples, and context.
That sounds obvious, but it means I am doing more setup work up front:
- writing clearer docs
- defining edge cases
- breaking things into smaller tasks
- being explicit about tradeoffs
This is good engineering practice anyway, but it is still work. It is not "free speed."
3) I write less code, but I read more code
This is probably the biggest shift for me.
AI can generate a lot of code quickly. That does not mean I can trust it quickly.
So I spend more time reviewing:
- correctness
- edge cases
- performance
- maintainability
- security
I type less. I read more. A lot more.
And reading bad or almost-right code for long stretches is a different kind of fatigue than writing code from scratch.
4) The judgment burden is higher, not lower
AI output can look clean and still be wrong.
Sometimes it compiles but misses the actual requirement.
Sometimes it follows the prompt but makes a bad architectural choice.
Sometimes it introduces subtle bugs that only show up later.
So the real bottleneck is not typing speed. It is judgment.
If anything, AI makes engineering judgment more important, not less.
Ownership does not change
This part is easy to forget.
AI can help me move faster, but it does not own the outcome. I do.
I still own:
- reliability
- maintainability
- security
- delivery quality
That part has not changed at all.
Where I am landing
I am not anti AI. I use it constantly and it is genuinely useful.
I just think the conversation is too focused on speed and not enough on the mental cost of working this way every day.
AI reduces some kinds of work, but it increases others. And some of the new work is more tiring than people expect.
It is still worth it for me. But the fatigue is real.