Toolkit

Generative AI Assistant & Agent Prioritization Canvas

An interactive prioritization canvas to capture, score, and pilot assistant- and agent-style generative AI opportunities. Includes fields for scope, inputs, outputs, data and safety guardrails, feasibility notes, a pilot-readiness checklist, a starter prompt library, and simple scoring guidance to help teams choose high-impact, low-risk pilots.

Interactive Tool

Generative AI Assistant & Agent Prioritization Canvas

How to use this canvas

Use this interactive canvas to capture a candidate assistant or agent opportunity, assess feasibility and risk, and decide whether to run a short proof-of-value pilot. Complete each field with practical specifics. The Pilot Readiness Checklist helps you confirm minimum safety, data, and measurement requirements before starting. Save the canvas so you can compare candidates later.

Quick scoring guidance

Rate Business impact and User value on a 1–5 scale (higher is better). Rate Risk on a 1–5 scale (higher is more risky). A simple priority heuristic: Priority indicator = (Business impact + User value) - Risk. Use the Priority rationale field to explain the result and recommended next step (pilot, research, or pause).

Describe the user, the context, the desired outcome, and why it matters. Example: 'Technical support rep needs a one-line summary of customer device state to resolve 70% of tickets faster.'
List exact inputs the agent will receive (user message, API data, files) and clearly state what the agent must not do (decision boundary).
What concrete outputs or actions should the agent produce? e.g., recommended reply, ticket update, data extraction, scheduling action. Be specific about format and required fields.
Choose the intended operating model for this opportunity.
Does this require workflows, multi-step agents, human approvals, cross-system transactions, or scheduling? List systems and APIs involved.
Define measurable metrics: time saved, error reduction, throughput, NPS, compliance rate. Include baseline values when possible.
List required data sources, data sensitivity (PII?), access levels, retention, anonymization, allowed uses, and logging/audit requirements.
Dependencies, available APIs, latency or compute constraints, model availability (LLM type), required training/fine-tuning data, and engineering effort.
Consider potential harms, bias, hallucination risk, legal/regulatory issues, user deception, and conditions that require immediate human intervention.
1.0 10.0
1.0 10.0
1.0 10.0
Synthesize scores and explain whether to run a short proof-of-value pilot (include proposed scope and primary metric), conduct more research, or pause.
Check items that are completed before starting a trial.
Seed the agent with 3–5 example system or user prompts. Use placeholders for variables (e.g., {{order_id}}). Example starter: 'System: You are an assistant that summarizes order status in one sentence. User: Provide order {{order_id}} status.'
Typical pilots last 2–8 weeks.
You can explore this tool now. Sign in or create an account to save your responses and return to them later.
Sign In to Save Create an Account
Make this tool part of your work

Save a personal copy, bring it to your team, or tailor the questions and workflow to fit what you are hungry to improve.

Member customization and team collaboration are coming soon.
Discuss Ask the Librarian

Discussion

Comments and conversation will live here.