Prompt Monarchism

From the AI Research Compass wiki

Prompt Monarchism: A single carefully written system prompt is enough to govern a model's behavior across every task; no further machinery is required.

Beliefs

Prompt Monarchism holds that the system prompt is the crown. One well-drafted block of instructions, placed at the top of the context, defines the model’s persona, rules, tone, refusals and task procedures, and the model obeys it everywhere. Workflows, routers, fine-tuning and external guardrails are seen as needless ministries: if the behavior is wrong, rewrite the decree.

It sits in the upper-left of the chart. The language lean is strong because governance happens purely through natural-language text. The Monolith lean comes from its reliance on one model plus one prompt, with no surrounding system. Its paradigm level is Mainstream: this was the default way to build on language models in the early chat era, and newer approaches to context and agent design define themselves partly in opposition to it.

Arguments

  • Instruction-tuned models follow detailed system prompts well enough that many applications ship with little else.
  • A single prompt is cheap to iterate, easy to version and readable by non-engineers.
  • In-context instructions adapt a model to new tasks instantly, without training or infrastructure.

Critiques

  • Long monolithic prompts become brittle: instructions conflict, get ignored deep in context and are vulnerable to prompt injection.
  • One static prompt cannot supply the task-specific data, tools and memory that complex jobs need.
  • Context-engineering rivals argue the prompt is just one small part of what the model sees, and rarely the most important.

Representative works

Relationships

How to get it

Land on this cell in the full or pro test: x ≈ −0.62, y ≈ +0.69, paradigm ≈ −0.58.

A first draft. Corrections and better arguments are welcome through a pull request. Edit this entry on GitHub Edit its data