Pattern: Agent as Drafter
🔵 Optional
Use this when: You need to produce a document, story, report, or any written artifact and want the agent to write the first draft while you direct and refine.
How it works
The agent drives (writes). You navigate (review, critique, refine).
You → describe what you need
Agent → produces a first draft
You → review critically, provide feedback
Agent → revises
You → review again, iterate until quality is right
You → final edit, then commit
This is the most common PM-agent interaction pattern. It works for user stories, PRDs, sprint reports, stakeholder updates, documentation, research plans, and any other written output.
When to use it
- Writing user stories with acceptance criteria
- Drafting PRDs or product briefs
- Creating sprint reports and stakeholder updates
- Writing research plans and discussion guides
- Documenting decisions and trade-offs
- Creating templates and checklists
When NOT to use it
- Strategic decisions — Don't ask the agent to decide your product strategy. It can draft options, but the decision is yours.
- Sensitive communications — If the message involves organizational politics, personnel issues, or bad news, write it yourself. The agent doesn't understand interpersonal dynamics.
- Creative vision — If you're defining a product vision or narrative, start with your own words. Use the agent to refine, not originate.
How to get good drafts
1. Provide rich context
Bad: "Write a user story for login."
Good: "Write a user story for account registration. The user is a small business owner who wants to try the product for the first time. They should be able to register with email and password. We need Gherkin acceptance criteria covering the happy path, validation errors, and duplicate email handling."
The more specific your input, the less revision you'll need.
2. Specify the format
Tell the agent exactly what structure you want. Include examples if possible.
Draft this in the following format:
- Title: As a [user], I want to [action] so that [benefit]
- Description: 2-3 sentences of context
- Acceptance criteria: Gherkin format (Given/When/Then)
- Out of scope: What this story does NOT include
3. Specify the audience
"Write this for our executive stakeholders who have 2 minutes to read it" produces very different output than "Write this for the engineering team who needs to implement it."
4. Ask for options
Instead of asking for one draft, ask for 2-3 approaches with different trade-offs. Then pick the best one and refine.
The revision cycle
The first draft is never the final product. Plan for 2-3 revision rounds:
Round 1: Accuracy and completeness
- Is anything factually wrong?
- Is anything missing?
- Is the scope right?
Round 2: Quality and clarity
- Is the language precise? (No "intuitive," "seamless," "robust")
- Is it the right length?
- Would the target audience understand it immediately?
Round 3: Your voice
- Does it sound like you?
- Is the tone appropriate for the audience?
- Would you be comfortable presenting this verbally?
Red flags in agent drafts
Watch for these common issues:
| Red flag | What's happening | What to do |
|---|---|---|
| Vague language ("ensure a seamless experience") | The agent is papering over a detail it doesn't know | Ask: "Be specific — what does this interaction actually look like?" |
| Comprehensive scope | The agent included everything it could think of | Cut aggressively — find the smallest valuable version |
| Confident claims without evidence | The agent is asserting things as fact | Ask: "What evidence supports this?" or add your own |
| Perfect first draft | Either the task was trivial or you're not reviewing critically | Challenge it: "What's wrong with this draft?" |
| Lists of 10+ items | The agent is being thorough, not prioritized | Ask: "Rank these by importance and cut the bottom half" |
AI in practice: drafter patterns across industries
The Agent as Drafter pattern is the most widely applicable AI workflow:
- Contract drafting and review — Legal teams draft contracts from templates and flag deviations from standard terms. Review time per contract drops from 4 hours to 45 minutes. See Content & Documentation use cases.
- Policy documentation with compliance review — AI drafts policy documents with inline legal commentary, streamlining multi-stakeholder review cycles. See Finance & Analysis use cases.
- RFP response acceleration — A consulting firm matches RFP questions to past responses, drafts answers, and flags gaps. Turnaround drops from 2 weeks to 3 days. See Sales & Marketing use cases.
Connecting to workflows
This pattern is used in: