continuationPreamble

function continuationPreamble(current: Intent, remainingMs: number): string

import { continuationPreamble } from '@driftengine/ai';

Parameters

In depth

Three lines. Without them the model answers as though acting immediately, proposes what is right now and wrong in a second and a half, and the lookahead becomes lag that is still paid for. The whole value of asking early is lost to not saying so.

What it costs: three lines of every continuation's context budget. What would make it wrong: a provider that ignores framing of this kind entirely, at which point the lead should shrink toward zero rather than the preamble being dropped — the fix for a model that cannot reason about the future is to stop asking it to.