While tweaking some log messages, I was reminded that context is everything.  An event without context is hard to act on.  That made me think about LLM prompts: the instruction is rarely the problem; the missing context is.

This isn't only about software, it's just a general rule.  Without context, information doesn't help much.

Keeping that parallel in mind when I'm building software or working with tools like LLMs helps me trust my intuition.

So instead of thinking "What should I log?" or "What should I paste into the prompt?", I ask: what's the smallest amount of information a human would need to take the right action for this event?

What parallels do you use when you're trying to improve how you interact with software?

devs

While tweaking some log messages, I was reminded that context is everything.  An event without context is hard to act on.  That made me think about LLM prompts: the instruction is rarely the problem; the missing context is.

This isn't only about software, it's just a general rule.  Without context, information doesn't help much.

Keeping that parallel in mind when I'm building software or working with tools like LLMs helps me trust my intuition.

So instead of thinking "What should I log?" or "What should I paste into the prompt?", I ask: what's the smallest amount of information a human would need to take the right action for this event?

It's been a useful mental model.

What parallels do you use when you're trying to improve how you interact with software?