On convenience and understanding

When we automate something we’re choosing to not have to think about the concerns we’re automating away. This is fine since we can’t know everything. It’s good to acknowledge this. The stuff a learner has to know about changes in an interesting way in that they still have to know about the thing they’re trying to setup plus additionally a bunch of things related to how the automation works. (And of course what do you do when the abstractions leak …)

Being intentional about friction

Thinking about how people get started with a new website has changed recently with clankers. On the table now is “from nothing, and prompt”. This skips a lot of learning as it turns out. Not all learning is super useful probably but some of it is. And not starting from zero can help somebody get their berrings, or avoid problems others have seen before. Starting from 0 does seem like a step … back / sideways? … too far.

  • LLMs make it easier to avoid learning, but also create new opportunities to learn.
  • LLMs enable more people to make simple apps, but complex systems still require developer skills.
  • Acquiring developer skills still involves learning to code.
  • LLMs open new entry paths for acquiring dev skills, especially with support from existing developers.
  • The productivity gains of LLM coding are exaggerated, and reduced or reversed with overreliance.
  • LLM code generation is not very useful on niche or novel problems.