Prompt Cards

Create a decision brief comparing vendor platforms vs internal build across control, speed, TCO, compliance, talent, and lock-in. Provide a hybrid option. Output: scorecard, phased migration path, and exit strategy clauses.

Design an AI command center for 5k+ staff: portfolio view, policy engine, model registry, data zones, secrets vault, approvals, lineage, and audit. Include risk taxonomy (legal, security, brand), RACI, escalation flow, and board-level KPI set.

Build a FinOps plan for AI: unit economics per task, cost alerts, usage quotas, fallback models, and A/B testing harness. Output dashboards (tasks/hour, cost/task, quality), run-book for anomalies, and a monthly optimization ritual.

Ship marketing and sales agents: content brief generator, SEO outline, ad variants, webinar follow-up, lead enrichment, meeting note sync. Define prompts, tool chain (CRM, MAP, docs), eval rubric (brand fit, factuality), and weekly growth experiments. KPIs: MQLs, reply rate, CAC.

Create a pragmatic policy: data map, lawful bases, retention, DLP, redaction, secrets handling, vendor assessment, human review. Include a DPIA lite template, audit log fields, and a quarterly red-team plan. Output training slides for managers and a 1-page employee FAQ.

Design an orchestration layer for 5–10 agents: routing, memory, tool auth, retries, tracing, and guardrails. Compare options (workflow engines vs. code), propose SLAs, queues, circuit breakers, and observability (logs, spans, evals). Output a reference architecture and a minimum viable mesh.

Act as program manager for a 200–800 staff company. Deliver a 30/60/90 plan: portfolio of 6 agents (sales, finance AP, IT help, HR Q&A, legal intake, PMO summarizer), governance (RBAC, data retention), platform selection, integration map, pilot design, scorecards, change-mgmt comms. KPIs by function. Budget tiers: $5k/$20k/$50k.

Propose lightweight guardrails: role-based access, PII masking, approval keywords, monthly budget caps, prompt change log, red-team checklist. Output SOPs (1 page each), incident playbook (who/when/how), and a weekly scorecard (usage, spend, errors, escalations).

Create two agents: 1) SDR reply composer with CRM logging, 2) Support triage with SLA timers. Define prompts, tool calls (email, calendar, CRM/helpdesk), do/don’t rules, tone, escalation, and weekly A/B tests. KPIs: first-response time, qualified replies, resolution rate, deflection %, NPS.

Design a data-lite RAG plan without engineers. Sources: Google Drive/Notion/PDFs. Deliver: folder taxonomy, redaction rules, ingestion checklist, embedding strategy, update cadence, eval set of 25 Q&A, precision/recall targets, “can’t answer → escalate” rule. Include cost model (<$150/mo) and privacy notes.

You are a product ops partner. Create a 1-page catalog of 10 agent ideas for a <50 staff firm: sales reply, invoice Q&A, website chat, calendar triage, hiring screener, policy Q&A, SOP generator, social captioner, post-meeting summarizer, vendor email parser. Score by impact, effort, risk. Output a top-3 with JTBD, prompt skeletons, tool permissions, fallback to human.

Act as an AI rollout lead for a sub-50 person company. Deliver a 14-day plan: use cases ranked by ROI/risk, 2 quick-win agents (inbox triage, FAQ/RAG), minimal governance (human-in-the-loop), success metrics (time saved, CSAT, lead response). Constraints: no dedicated ML team, shared SaaS tools. Output: day-by-day schedule, tools list, prompts, guardrails, pilot script, acceptance criteria, retro template. Keep cost < $1k, setup < 8 hours.