NT
AboutInterestsProjectsBookshelfEssaysHire Me

LLMs are for grunt work

The wrong way to use AI is to substitute it for your own thinking.

The right way to use AI is to substitute it for all "trivial thinking."

For any given skill level someone obtains, all the different tasks to which that skill can be practically applied fall into three buckets:

  • too challenging: you don't yet know how to solve the task so you can't even get started
  • just right: you have just the right skill level to solve the task, it's challenging enough to be fun and exciting
  • trivial: the task is too easy, you've successfully completed so many tasks of this type for there to be zero surprises. Boring work falls here.

Using AI to do all the things you don't understand means that you'll never understand them. Sometimes that's okay, but the missed learning should be considered and fully priced as a trade-off.

Using AI to do the things you’d otherwise find exciting and engaging is perhaps the biggest faustian bargain - you miss the most potent and enjoyable learning and self-development opportunity. If this becomes a habit, you’ll simply stop developing and life becomes much less enjoyable.

However, using AI for trivial work is perfect. Without LLMs, our work fluctuates from fun to boring all the time. When we’re having fun, we stay in a pleasant flow state - keeping things both productive and enjoyable. Trivial work can easily break that flow. By handing trivial work to LLMs, you can spend more time working in perfect balance with challenge and ability.

Things I always hand off to LLMs:

  • finding sources when doing research BUT NOT doing the analysis and synthesis
  • writing boring administrative emails BUT NOT creative writing
  • implementing simple software components BUT NOT architecting solutions
  • processing/sorting/formatting my rough notes BUT NOT thinking through problems

As for the things that are "too challenging," I use LLMs to break them down in terms of what I already understand, which puts them into the realm of what's fun and exciting to solve.

In short: give clankers the grunt work but never stop using your brain.


other essays:

Small Teams
Aug 2025
How I read
Mar 2024