For me the best part about AI is that when I'm feeling lazy, I can tell the AI to do it. Whether it gives me gold or gives me shit, it doesn't matter, because I have now started my work.
For example, the other day I needed to find which hook to use in a 2300 loc file of hooks. AI found me the hook in 5 seconds and showed me how to use it. This is a pure win - now that I have the name of the hook I can go read it in 30 seconds and verify it's what I wanted. If it's not, I can ask again. There's zero risk here.
Re: usage of LLMs, that is honestly the way I like to use LLMs. That and auto-complete.
It's great for creating a bird's-eye view of a project because that is very fuzzy and no granular details are needed yet. And it's great at being fancy autocomplete, with its stochastic bones. But the middle part where all the complexity and edge cases are is where LLMs still fail a lot. I shudder for the teams that have to PR review devs that jubilantly declare they have "5x'ed" their output with LLMs, senior or not.
What is even more worrisome is that the brain is a muscle. We have to exercise it with thinking, hence why puzzlers stay sharp at old age. The more you outsource your code (or creative writing) thinking, the worse you get at it, and the more your brain atrophies. You're already seeing it with Claude Code, where devs panic when they hit the limit because they just might have to code unassisted.
Lesson learned: being lazy with AI has a hidden cost.
On small issues it's usually just a glance over the diff, on harder issues it's not challenging to scold it into a good path once something starts coming together, as long as it's a localized edit with a problem I understand. I'll often take over when it's 40-60% there, which is only possible b/c it does a good job of following todo lists.
I've moved into a good daily workflow that emphasizes my own crafting during the times I'm sharp like mornings, and lets AI do the overtime and grunt work while I prepare for tomorrow or do more thoughtful writing and design.
I spend probably 70% of my active coding time coding. The rest is using LLMs to keep making progress but give my brain a bit of a break while I recharge, being more supervisory with what's happening, and then taking over if necessary or after the 10-15 minutes it takes for me to start caring again.
It gave me a shit plan today. Basically I asked it to refactor a block of prototype code that consisted of 40 files or so. It went the wrong way and tried to demolish it from the bottom up instead of making it 100% backwards compatible. If it made a mistake we would have taken forever to debug it.
But yeah, it gave me something to attack. And the plan was fixed within an hour. If I tried to make one myself, I would have frozen from the complexity or gone in circles documenting it.
Having the AI just getting the wheels turning when I'm not in the mood myself has many times been a good way to make progress.
The biased behaviour towards just asking for a solution both devalues you and leads to any innovative outcomes disappearing before they had a chance to exist.
I've ended up being the company ideas man not because I'm good but everyone else stopped thinking.
They aren't the best solution for everything. Thinking is one of those things.
I do control systems that have to work first try and after running the plant through its paces I’m out the door never to return
But! There are some problems that are intrinsically interesting to me and others that are boring as hell but need to be solved.
In 20 years of work I’ve not come up with any way to make those boring tasks psychologically _not_ so disagreeable they are close to painfull.
But now the LLM can do whatever crap it does for those and I _finally_ can focus just on the interesting bits.
So, a boring task - off to the LLM you go - interesting - it’s mine, all mine!
Yes! And hell, checking out how the machine solved a boring thing can still be a source of interest in itself, so the dopamine squeezing novelty keeps on flowing all day long.
The best use cases for me have been when I know the solution and what it should look like but don't want to write it all by hand myself. The AI can make the code just appear and I can verify it is up to spec.
Does the answer change if the author is an LLM as opposed to another person?
Climb down from your high horse why don't you.
If your coworkers can get into the groove faster than you (for whatever reason), you’ll continue to do quality work, not look busy, and no longer be paid for it.
And good luck finding anyone these days who can do quality.
It's a nice niche.
I don't mean to attack this person specifically but it's a frankly pretty bad mindset that is far too prevalent in our profession - the way I do things in the Right Way, I write Good Code, very few other people write Good Code, and because I do things the way they were done 5, 10, 25 years ago means that my way is better than yours.