In the fast‑moving world of generative AI, prompt engineering is equal parts art and science. Traditional workflows rely on a single author iterating in isolation, producing slow feedback loops and narrow perspectives [1]. We propose Collective Prompts: a real‑time, agent‑assisted framework that enables teams of humans and autonomous agents to create, refine and validate prompts collaboratively.
A core strength is participation flexibility—prompt generation may be human‑only, agent‑only, or a hybrid. Whether a brainstorming room for people or a swarm of specialised evaluators running at machine speed, Collective Prompts adapts to its users’ context.
By allowing parallel contributions, the system orchestrates a living, adaptive dialogue. Each prompt becomes a shared artefact—continuously enriched by comments, challenges and new angles—while agents handle tasks such as fact‑checking, clarity rewrites and ethics vetting.
Contributions flow into a shared refinement loop. Every variant is tested in a sandboxed RAG environment that measures retrieval performance and generation quality . A consensus engine ranks candidates using objective metrics and subjective votes.
The framework therefore aims not only to accelerate prompt engineering, but also to make it measurable and collaborative—bridging human intention and machine execution through an orchestrated cycle of generation, evaluation and reinforcement.