Step 2 of 3 - Requirements clarification
Clarify What You Want
Turn Phase 1 evidence into confirmed future requirements without designing the assistant yet.
- You need
- The complete Phase 1 report and access to the primary user for a focused conversation
- Keep for Step 3
- A Phase 2 requirements clarification report
What this step does
Requirements clarification validates important observed patterns, identifies material decisions the evidence cannot answer, separates preferences from workarounds, and records explicit future needs. It does not select integrations, create workflows, write instructions, or design the assistant.
Start with the complete Phase 1 report. Ask the primary user only the highest-value unanswered questions.
What to prepare
- The full Phase 1 work-pattern evidence report, including source gaps and confidence.
- Access to the primary user for a focused, one-question-at-a-time conversation.
- A safe place to retain the Phase 2 clarification report for instruction design.
Copyable Prompt 2 - Requirements Clarification Conversation
Paste this into the same AI conversation after completing Phase 1. It creates the questionnaire dynamically rather than asking every user the same fixed list.
I am continuing requirements discovery for a future personal work AI assistant for [PRIMARY USER]. This is Phase 2: Requirements Clarification. You have completed Phase 1: Work Pattern Evidence Collection. Use the full Phase 1 report as the primary input. Your job is to identify the material design decisions Phase 1 cannot answer, validate important patterns, capture explicit future needs, and create a requirements record. Do not design the assistant or write its instructions yet. Before each question, identify what Phase 1 established, the unresolved design decision, and the minimum information needed to resolve it. Do not ask a question that high-confidence evidence already answers. Ask one primary question at a time. Ask a brief follow-up only when needed. Classify each validated pattern as Confirmed (desired future behavior), Contextual (applies only in stated conditions), Legacy (an old practice), Workaround (caused by inadequate information/systems), or Incorrect Observation. Treat direct future needs as New Explicit Requirements. If a stated preference differs from observed behavior, do not choose silently or argue. State the observed pattern briefly and ask what the assistant should do in the future. Record the observed behavior, stated preference, and resolved requirement separately. Target 10–20 high-value primary questions, prioritizing frequency, business value, risk if wrong, ambiguity, and user impact. Stop once material questions are resolved; do not over-interview. Prioritize These Design Areas When Evidence Makes Them Relevant Briefing and information consumption What should appear first; detail level; when decision, risk, owner, date, financial impact, or evidence should surface; preferred rapid-consumption format. Role-specific intelligence Which work objects matter to this user - for example customers, projects, systems, vendors, contracts, products, operations, or finances - and what belongs in a default brief versus on request. Meeting preparation Which meetings deserve preparation; timing; what prior context, commitments, decisions, questions, and follow-up should appear. Decision support When options versus recommendations are wanted; evidence threshold; confidence; tradeoffs; financial and risk treatment; prior decisions. Proactive behavior What is an immediate alert, next briefing, periodic summary, only-on-request item, or noise. Commitments and follow-up What to track, remind, flag, draft, monitor, or escalate; what always requires approval. Challenge history trust and authority When to flag conflicts; history relevance; source freshness; missing/conflicting data; confidentiality; what the assistant may answer, analyze, recommend, prepare, draft, monitor, or act on. Interview Rules Use a concrete tradeoff or scenario when it produces a clearer answer than an abstract question. Pay close attention to workarounds. Repeated questions may show a visibility gap, not a desired default behavior. Keep a Requirements Clarification Record: question ID, design area, Phase 1 evidence, question, answer, follow-up, final interpretation, status, and confidence. Ask remaining high-impact questions before ending. Do not continue solely for completeness. Required Output After the Conversation Confirmed operating preferences for briefings, detail, decisions, risk, financial information, dates, ownership, and history. Role-specific intelligence requirements, using Always Include, Include When Relevant, Only When Requested, Not Needed, and Unresolved classifications. Meeting preparation, decision support, proactive intelligence, commitment/follow-up, challenge/pushback, trust, confidentiality, and authority requirements. Workaround findings, stated-preference-versus-observed-behavior resolutions, new explicit requirements, and only the remaining material unknowns. A Consolidated Requirements Dataset table with: Requirement ID; domain; requirement; source; trigger/situation; expected behavior; priority; confidence; evidence/rationale; and open design question, if any. Strict Boundary Do not design the agent, select integrations, create workflows, or write instructions in this phase. Begin by reviewing Phase 1, then ask the single highest-value unanswered question.
Make it yours
You may adjust the role, work domains, approved sources, analysis period, output length, and organization-specific authority limits. Do not bypass evidence collection or confirmation: the Phase 1 report is what keeps the conversation grounded in actual work.
What happens next
Retain the full Phase 2 clarification report. It feeds Step 3, where confirmed requirements become assistant instructions, data requirements, action boundaries, and traceability.