Intent Context

Give the model the why, not just the what. Intent context is the practice of supplying the purpose behind a request — the larger task, who the output is for, what it needs to enable — instead of (or before) piling on step-by-step instructions. A capable model that understands your intent can connect the task to the right information and infer the unstated details, rather than guessing what you meant and defaulting to a generic response. The source frames it as "context over detail": one line of "here is why I am asking" does more than a long procedural spec. It is the highest-signal thing you can add to a prompt precisely because it is the one thing the model cannot derive on its own.

Concretely, instead of "Write me an email to a client about the delay," supply "I'm working on [the bigger task] for [who it's for]. They need [what the output enables]. With that in mind: [your request]." When the operator has a durable Context Substrate (second brain / AIOS), stating intent also cues the model to pull the right context files.

Claims

A model that understands your intent connects the task to the right information instead of guessing — supplying the "why" is higher-leverage than adding procedural detail. principle — durable: intent is the one input the model cannot reconstruct from the task text alone, so it carries the most signal per token.
Prefer intent/context over step-by-step detail when prompting a capable model. best practice — context: strong instruction-following, reasoning-capable models (the source attributes this to Anthropic's Fable 5 guidance); on weaker or more literal models, explicit procedure may still earn its place, so "context over detail" is contingent on model capability.
Adding the "why" is the part of a prompt to keep even while cutting the rulebook — it is real signal, not redundant procedure. best practice — context: shortening over-detailed prompts (see Concise Prompting); the reconciliation is "add the why, cut the rules," so intent context is what survives the trim.

Concise Prompting — the complement: keep the why (intent), cut the redundant rulebook. Together they are "say less, but say the right thing."
Negative Prompting — intent tells the model where you're headed; negative prompting tells it where to stop. Both are lighter-touch steering for a capable model.
Context Substrate — a durable context layer is where the "why" often already lives; stating intent cues retrieval of the right context files.
Agentic Simplicity — "smarter model, lighter touch": give intent, not an exhaustive spec.
Meta-Prompt — a reusable prompt template can bake an intent slot ("here is why I am asking") into its shape.
Distillate: How Anthropic Engineers Actually Prompt Fable 5 — six habits for a smarter, lighter touch

Intent Context

Claims

Related

Linked from