Playbook: Design High-Value Prompts, Assistants, and Workflows

Preview Cards

Here are the first 5 questions. Create a conversation to invite someone and discuss each card.

1. <section> <h2>Welcome — Make AI assistants you can trust and reuse</h2> <p>Teams waste time on endless prompt fiddling and brittle assistants because they start with vague goals instead of testable behaviors. This playbook helps you design prompts, personas, and workflows that reliably produce the outputs you need — and that you can test, iterate, and maintain.</p> <h3>Who this helps</h3> <ul> <li>Product managers and team leads who need repeatable assistant behavior without becoming prompt engineers.</li> <li>Knowledge workers (analysts, educators, clinicians) who want clear, defensible outputs.</li> <li>Small teams and non-technical staff who want ready-to-use templates and tests.</li> </ul> <h3>What you’ll get</h3> <ol> <li>A practical guide to writing high-value prompts that reduce ambiguity and failure.</li> <li>Persona templates you can copy and adapt for common roles.</li> <li>Patterns for chaining model calls and integrating retrieval or tools.</li> <li>A concise testing checklist and an interactive test form to capture experiments.</li> </ol> <h3>How to begin</h3> <p>Start with one small, important workflow: the task you want the assistant to do reliably today (e.g., triage incoming requests, summarize clinical notes, draft a customer reply). Use the guide to draft a system prompt and 3–6 user examples, then run the test form to capture results. Iterate until the assistant meets minimum quality on the checklist — then expand scope.</p> <p><strong>Quick promise:</strong> leave this resource with a working prompt + persona + one repeatable test that your team can run and refine.</p> </section>
2. <section> <h2>Guide: Design High-Value Prompts — principles, patterns, and examples</h2> <p>Good prompts do three things: they (1) reduce ambiguity about intent, (2) provide relevant context or constraints, and (3) define what success looks like. This guide turns those principles into patterns you can apply across roles.</p> <h3>Core principles</h3> <ul> <li><strong>Be specific about role and goal:</strong> Tell the model who it should be and what outcome you expect (not just the task).</li> <li><strong>Provide structured context:</strong> Include data fragments, relevant facts, or retrieval cues so the model needn't guess.</li> <li><strong>Define constraints and format:</strong> Specify length, tone, forbidden outputs, output schema (JSON, bullet list), and when to ask clarifying questions.</li> <li><strong>Use examples and edge cases:</strong> Few-shot examples anchor behavior; include an example of a bad answer plus a corrected one if useful.</li> <li><strong>Test for robustness:</strong> Create a small suite of tests covering typical, ambiguous, and adversarial inputs.</li> </ul> <h3>Prompt-building pattern</h3> <ol> <li><strong>System instruction:</strong> A short role-setting sentence that persists across the interaction. Example: "You are a concise customer-support agent who cites policy sections and suggests next steps."</li> <li><strong>Context:</strong> Facts, retrieved documents, or dataset crumbs. Use clear labels: "<em>Policy:</em> ..." or "<em>Customer message:</em> ..."</li> <li><strong>Task:</strong> One-sentence instruction that starts with an action verb: "Summarize, classify, generate, compare, propose."</li> <li><strong>Constraints & format:</strong> E.g., "Respond in 3 bullets. Each bullet max 140 characters. Include one recommended action and a confidence score (1–5)."</li> <li><strong>Examples:</strong> 1–3 user→assistant pairs demonstrating the desired output.</li> </ol> <h3>Common prompt patterns</h3> <ul> <li><strong>Instruction-with-format:</strong> Best for predictable outputs (forms, email drafts, checklists).</li> <li><strong>Chain-of-thought selective:</strong> Ask for brief reasoning only in tests to improve evaluation, but avoid exposing lengthy internal chains in production unless needed and controlled.</li> <li><strong>Retrieval-augmented:</strong> Provide retrieved passages and instruct the assistant to cite sources or refuse if no relevant passage exists.</li> <li><strong>Classification → Generation:</strong> First ask the model to classify or extract structured fields, then feed results into a generator to produce the final response.</li> </ul> <h3>Three short examples across audiences</h3> <h4>Non-technical product manager — Feature summary</h4> <p>System: "You are a concise product writer who uses plain language and highlights user impact."<br/>Task: "Summarize this user interview into 4 bullets: problem, frequency, impact, and a proposed experiment. Bullets max 120 characters."</p> <h4>Healthcare intake assistant — Triage note</h4> <p>System: "You are a clinical triage assistant. Never give medical advice; only suggest next steps and escalate criteria."<br/>Task: "From this intake form, extract symptoms, duration, and red flags. If red flags present, recommend urgent clinician contact."</p> <h4>Skilled trades — Repair estimate draft</h4> <p>System: "You are an experienced HVAC technician who writes clear estimates."<br/>Task: "Draft a one-paragraph repair estimate from this job description and list three parts likely needed (name and estimated price)."</p> <h3>Practical tips to reduce brittleness</h3> <ul> <li>Keep the system instruction stable; don’t hide critical constraints in ephemeral user messages.</li> <li>Prefer precise output schemas (JSON keys) when the output feeds automation or records.</li> <li>Fail explicitly: instruct the assistant to answer "I don’t know enough" or ask for clarification rather than guess.</li> <li>Limit creativity for operational tasks by tightening constraints (format + examples + refusal conditions).</li> </ul> <h3>Next steps</h3> <p>Use the persona templates in the reference, then run tests from the checklist. Record results in the interactive test form to create regression tests and track improvements over time.</p> </section>