AI prompt for agile stories

gistfile1.txt

Most teams use AI for code generation. We use it to force product discipline.

**The Problem:**
Vague epics → ambiguous features → confusing stories → engineering waste.

When a PM says "reduce cart abandonment" without metrics, acceptance criteria, or clear success measures, engineers spend 2 weeks building the wrong thing. Backlog refinement becomes a bottleneck.

**The Solution:**
Structured prompts to AI agents that enforce rigor before code is written.

Instead of hoping a PM nails the specification, we give Copilot agents (or Claude Code Agents, or Gemini Gems) explicit templates:

✓ **Epic Agent**: Turns business goals into structured epics with:
  - Business value in *metrics*, not feels ("65% cart abandonment → target 42% completion")
  - 4-7 distinct features that can ship independently  
  - Success metrics that measure *outcomes*, not outputs
  - Explicit risks, constraints, and what's OUT of scope

✓ **Feature Agent**: Breaks epics into shippable features with:
  - User story format (As a X, I want Y, so that Z)
  - Testable acceptance criteria (Given/When/Then format)
  - Design links + technical considerations
  - Clear success metrics (not "customers are happy" but "30%+ click rate")

✓ **Backlog Item Agent**: Creates sprint-sized stories with:
  - Definition of Ready checklist (before eng starts)
  - Definition of Done checklist (before shipped)
  - Clear dependencies and effort estimates
  - Acceptance criteria QA can't misinterpret

**Real Example:**

PM Input: "We need to reduce cart abandonment. Show customers delivery dates."

Epic Agent Generated:
```
Epic: Reduce cart abandonment through transparent delivery time visibility

Business Value: 65% abandonment + competitor shows dates → 8-12% improvement if we do too
  Saves ~$2M/quarter in lost revenue

Scope: Delivery date API, cart display, shipping method selector, A/B test
Success Metrics: 42%+ checkout completion (vs 35%), 40%+ users select paid express

Timeline: Q1 (3-4 sprints) | Risks: Carrier API latency | Out of Scope: International
```

Feature Agent Generated (for first feature):
```
Feature: Display estimated delivery date in cart review

Acceptance Criteria:
AC1: Date updates when shipping method changes (no reload, <500ms)
AC2: Show date in friendly format ("Arrives by Friday, March 5")
AC3: Fallback if carrier API down ("Arrives in 3-5 business days")
AC4: Mobile responsive (>14px font), accessible to screen readers

Success Metrics: 75% reach rate (customers see the date)
```

Backlog Items Generated (sprint-sized stories):
```
Story 1 (Backend): Build /shipping/estimate API endpoint [M - 3-5 days]
Story 2 (Frontend): Display delivery date on cart page [M - 3-5 days]
Story 3 (QA): Test edge cases + performance [S - 1-2 days]
```

**The Real Win:**
The PM still makes all decisions. But now:
- Engineers get unambiguous requirements, not guesses
- QA has testable criteria, not vague expectations
- Stakeholders see metrics upfront, not surprises after launch
- The backlog is disciplined enough that stories fit in sprints

The agent doesn't replace the PM. It enforces the discipline that good PMs do anyway—just faster, consistently, and at scale.

**For your team:**
- What's your biggest backlog bottleneck? Vague epics? Incomplete features? Ambiguous AC?
- Where could structured prompts save your team 5+ hours per week?

I'm sharing the actual agent prompts we use—if you want to see the full templates (Epic, Feature, Backlog Item agents with examples), DM me or comment below.

#ProductManagement #AI #Agents #Backlog #ProductStrategy #Copilot