The model only knows what's in the context window
A model has no memory of your team, your codebase, your writing voice, or the meeting you had this morning, unless you put it in front of it. It also has no way to ask a clarifying question unless you invite it to. This makes a vague request one of the most common causes of a bad AI response — not because the model is incapable, but because it was never given enough to work with, and it will fill the gap with a plausible-sounding guess rather than an admission that it doesn't know. Treating the model like a capable stranger who just joined the project — smart, widely read, but with zero context on your specifics — is a good working mental model.
A well-described task usually includes four things: the actual goal (not just the surface action — 'summarize this' is weaker than 'summarize this for a reader deciding whether to attend'), relevant context (constraints, audience, prior decisions, format expectations), explicit boundaries (length, tone, what to avoid, what NOT to change), and, where it helps, an example of the kind of output you want. You don't need all four every time, but omitting all four is how you end up with generic filler and then blame the model for being generic.