Skip to main content

The difference between context and prompt

There is a confusion we need to resolve: the confusion between prompt and context. A prompt is a point-in-time instruction. Context is a system composed of Agents that orchestrate, Rules that govern, and Skills that define concrete work. It's not an instruction — it's an information architecture.

DimensionPromptContext (Agents + Rules + Skills)
ScopeOne interaction, one task.The entire project. Multiple Agents, cross-cutting Rules, dozens of coordinated Skills.
CoherenceEach prompt is independent.Rules guarantee coherence across all Agents and all Skills.
IdentityAI has no role.Each Agent has identity, responsibilities, and limits.
StructureFree text.Architectural system: Agents with identity, Rules with hierarchy, Skills with dependencies.
EvolutionIt's consumed and disappears.Skills are created and completed; Rules evolve; Agents are calibrated.

Why this distinction matters

The confusion between prompt and context is not semantic. It's the difference between giving someone a one-time instruction and building a system that produces consistent results over time.

A prompt works for isolated tasks: "summarize this document," "write this email," "generate this image." The moment you need multiple interactions that must be coherent with each other — that must share conventions, respect constraints, and produce outputs that work as a system — a prompt is insufficient.

A project that uses prompts instead of context produces outputs that are individually correct and collectively incoherent. Each interaction makes locally optimal decisions that are globally contradictory. This is not a failure of the AI. It's a failure of the context.

Anti-pattern: The Context-Prompt

What it is: A long document without structure that gets copied into every interaction. There are no differentiated Agents, no explicit Rules, no self-contained Skills. It looks like context but behaves like a very long prompt.

What it produces: Without Rules, each interaction makes inconsistent decisions. Without defined Skills, tasks blend together. Without differentiated Agents, AI tries to be everything at once.

Quick diagnosis: If your context can't be separated into "Rules that apply to everything," "Agents that execute," and "Skills that define tasks," it's a long prompt, not a context.