{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreidviyzmgssaqtaqi5xfzf5mlikl7gst5rwwmlcldztp3kbtp5pkju",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mjyvjgwu2bo2"
  },
  "path": "/t/from-prompting-to-system-design-a-10-stage-model-of-llm-users/1379453#post_1",
  "publishedAt": "2026-04-21T10:14:53.000Z",
  "site": "https://community.openai.com",
  "textContent": "The 10-Stage LLM User Maturity Model\nStage 1 — Reactive User\n\nThe user immediately asks whatever comes to mind.\nThe focus is on receiving an answer.\nThe input is almost unchanged from raw language and has little structure.\n\nCharacteristics:\n\n“What is this?”\n“Explain this.”\n“Summarize this.”\n\nAt this stage, the user treats the model similarly to a search engine.\n\nStage 2 — Request-Oriented User\n\nThe user begins to understand the desired output format.\nThey can request summaries, tables, comparisons, or examples.\nHowever, they still do not directly influence the reasoning structure.\n\nCharacteristics:\n\n“Summarize in a table”\n“Compare pros and cons”\n“Include examples”\n\nThis is where output format control begins.\n\nStage 3 — Constraint-Oriented User\n\nThe user can apply constraints such as length, style, scope, and restrictions.\nThey intuitively understand that poor output is often due to insufficient constraints.\n\nCharacteristics:\n\n“Explain simply”\n“Avoid technical terms”\n“Only give three points”\n“Separate assumptions from facts”\n\nAt this stage, the input becomes a constrained task rather than a simple question.\n\nNote:\nEarly stages (1–3) often appear mixed in practice.\nA single prompt may contain elements from multiple stages simultaneously.\n\nStage 4 — Structured User\n\nThe user decomposes the question into components.\nThey separately define goal, scope, constraints, and output format.\nThey understand that model performance is strongly influenced by input structure.\n\nCharacteristics:\n\nGoal / Scope / Constraints / Output\nStep-by-step requests\nDefine first, then compare\n\nThis is the practical starting point of prompt engineering.\n\nStage 5 — Verification-Oriented User\n\nThe user does not simply accept answers.\nThey request separation of facts, assumptions, and uncertainty.\nThey consider hallucination, inconsistency, and model limitations.\n\nCharacteristics:\n\n“Separate evidence from assumptions”\n“Only state what is certain”\n“Say ‘unknown’ if unsure”\n“Check for contradictions with previous answers”\n\nAt this stage, the user sees the model as a fallible reasoning system, not just an output generator.\n\nStage 6 — Framing User\n\nThe user applies a specific analytical framework.\nThey prioritize how the question is processed over what is being asked.\n\nCharacteristics:\n\n“Focus on structure before meaning”\n“Analyze by cause / effect / impact”\n“Organize by pros / cons / risks”\n“Break down into structure / flow / importance”\n\nCore idea:\nThe user defines how to analyze, not just what to ask.\n\nStage 7 — Loop Design User\n\nThe user moves beyond single prompts and creates iterative structures.\nThey design loops of input → analysis → validation → refinement → feedback.\nOutput formats are treated as repeatable protocols.\n\nCharacteristics:\nAt this stage, the user treats the model not as a one-time tool,\nbut as part of a repeatable operational system.\n\nStage 8 — Meta Design User\n\nThe user moves beyond prompts and designs external structures.\nThey understand that the model is a black box and build meta-layers around it.\nAt this point, the user becomes closer to a system designer.\n\nCharacteristics:\n\nStructural routing\nExternal logging and validation flows\n\nAt this stage, the user is no longer just asking questions,\nbut orchestrating the system externally.\n\nStage 9 — Engine Orchestration User\n\nGiven the constraint that the model itself cannot be modified,\nthe user operates it as a system by generating, comparing, selecting, and refining multiple responses.\nThe focus shifts from a single answer to a structured process of candidate generation → evaluation → optimal selection.\n\nCharacteristics:\n\nGenerate multiple responses and compare them\nSelect and refine the most appropriate result\nEmphasize consistency and reproducibility\n\nAt this stage, the model is treated not as a conversational partner,\nbut as a controllable reasoning engine.\n\nThis transition occurs when the goal shifts from producing better answers\nto designing systems that consistently produce good outcomes.\n\nStage 10 — Interaction and Collaboration Design\n\nThis is the highest stage.\nHere, the focus moves beyond individual prompts to designing how humans and AI collaborate.\nIt is not about generating answers, but about building continuously improving interaction systems.\n\nCharacteristics:\n\nDefine roles between the user and the AI\nDesign collaborative structures\nBuild iterative and improving workflows\nStructure long-term problem-solving processes",
  "title": "From Prompting to System Design: A 10-Stage Model of LLM Users"
}