While I've found the inline recommendations or suggestions from both Copilot & Cursor to generally be good and helpful. It's made me ponder how to also get better high level suggestions.

function calcOpposite(foo: Foo): Foo

An assistant can most likely grab some information about what a Foo is here. But is the idea of its opposite too abstract? What sort of clues would be useful here?

If I'm making a switch statement of cases, would having some context give it the ability to more completely predict my goals? It feels obvious that this would be the case, and ironically got me thinking about documentation. While I try to write code dumb enough to not need any documentation, and recoil against unnecessary docs. It seems obvious that having something for the agent to reference is the solution.

switch (status): Color {
    case ...
}

As an experiment I'm going to try adding a SPECIFICATION.md document to some of my projects as I go. Dropping in bits of information that either already lives in external documents or as random facts in someones brain. Then hold thumbs our little assistant friends take the hint.