๐ Agent Product Requirements Document (PRD)#
A. Project/Agent Name:
B. Problem Statement:
C. Target Users / Stakeholders:
D. User Stories / Sample Prompts:
E. Key Capabilities (MVP):
Mark must-have and nice-to-have. Use checkboxes.
Natural language Q&A
Retrieval from internal docs (RAG)
Calls external APIs (specify which): ___________
Handles file inputs (PDF, images)
Follows multi-turn conversations
Handoffs/escalations to humans
Structured outputs (JSON)
Multimodal (image, audio)
Other: ___________
F. Guardrails / Risks / Constraints:
List safety, compliance, privacy, refusal rules, etc.
PII detection/redaction
Moderation API on inputs/outputs
Human-in-the-loop for edge cases
Rate limiting / cost cap
Blocklist for prohibited topics
Others: ___________
G. Success Metrics / KPIs:
How will you know this agent is successful?
e.g. 90% correct answers, <10% human handoff, sub-5s latency, user satisfaction, etc.
H. Evaluation Plan:
How will you test & validate agent performance? (Eval set, manual review, etc.)
I. Initial System/Assistant Instructions:
Example โsystemโ prompt for your agent. Try to make it realistic.
๐งฎ LLM Model Selection Worksheet#
Criteria |
Option 1 (e.g. GPT-4o) |
Option 2 (e.g. o3-mini) |
Notes |
---|---|---|---|
Context length needed |
|||
Speed/latency |
|||
Cost |
|||
Required features |
(function calling, vision?) |
||
Accuracy/quality |
|||
Compatibility (APIs/tools) |
|||
Team preference |
Decision:
Which model will you use for MVP and why?
๐งช Prompt Engineering Starter Table#
Prompt Version |
System/Assistant Prompt |
User Input |
Output/Goal |
Notes/Learnings |
---|---|---|---|---|
v1 |
||||
v2 |
||||
v3 |
โ Day 1 Success Checklist#
Agent PRD filled out and saved/shared
โFirst draftโ system/assistant prompt written
Three+ realistic user stories or prompts
LLM model selected (with rationale)
MVP features sketched (tools, RAG, outputs)
At least 2 guardrails/risks defined
Acceptance criteria/KPIs written
(Stretch) Run a test in Quickstart notebook