11–17 of 17 posts
The problem isn't "skill atrophy" — it's which skills matter more in a changing world. If LLMs genuinely do something better, let them do it and focus on what they do badly. What I'm exploring now: 1. How to convert tokens to value more efficiently 2. How to orchestrate a large LLM team instead of babysitting one session 3. How to parallelize work and make sure nothing blocks others 4. How to accelerate both producti…
A lot of the people I know who are just a few years younger than me learned / are learning at university and even grade school with the use of LLMs. This, to me, is terrifying. They are forming a foundation of knowledge and skills so fundamentally different to that of people my age and older.
This is not the difference between googling vs. finding a book in the library. This is the difference between being forced to think through a question and answer vs. not thinking through it at all.
I fear that the prioritization of LLMs in so many aspects of society and the workplace encourage this, and that we don't yet have enough guardrails to protect ourselves against this atrophy.
That is a long way to say, I also tend to fall on the side of caution. I do my best to use LLMs only for things that I can not do AND do not feel that I am likely to need to be able to do in the future. Or for things that I am super confident in my ability to do and thus know right from wrong and can just use LLMs to speed myself up.
I think that as long as people have a strong foundation (though it is unclear in what is important..) than using LLMs is not the end of the world. I am more worried about a future in which we won't have a useful BS meter to check them with.
I use LLMs as a coding assistant. I don't vibecode. The result is that I'm actually improving my skills much faster and better than before, while also working far more productive.
You practice doing challenging things yourself. Self-sufficiency is the name of the game. If you have already delegated your future to LLMs you are fucked and your skills will atrophy.
The best balance is give tasks to LLMs that consume large amounts of time but are either not challenging or heavily repetitive. My example would be handing off writing test automation cases but retaining all original authoring for experiments and design decisions.
I think about it differently. in this era you can pick up any new skill 10x faster than before if you have the right approach. so the hard skills themselves matter less now.
what actually matters is the stuff AI can't replace — your judgment, your ability to think critically, first principles reasoning. these are what make the final call on whether AI's output is right or wrong. because at the end of
the day, you're still the one making the decision.
so my approach: let AI handle the execution (coding, writing slides, whatever), but train yourself to evaluate whether what it produced actually makes sense. the skill that matters isn't "can I write this code" anymore — it's "can
I tell if this code is correct."
I've stopped worrying about learning specific programming languages or tools. those feel like the wrong things to invest time in now.