How GenAI helps engineers write better stories
In smaller engineering teams, developers often find themselves writing user stories without formal training in practice. The transition from understanding technical requirements to expressing them as well-structured stories presents specific challenges, particularly around vocabulary and decomposition techniques.
Consider this typical epic:
**Epic 1: User Auth & Profiles for marketplace**
*Implement user registration, login, authentication, session handling, and basic profile management.*
- User registration & login (user+ seller)
- Authentication (email/password)
- Session management
- Basic user profiles (name, email, preferences)
This epic bundle multiple concerns: different user types, frontend and backend components, and various system behaviors. Breaking it into implementable stories requires both technical understanding and knowledge of story decomposition patterns.
GenAI tools offer a practical approach to this challenge. Rather than replacing the engineer's judgment, they serve as collaborative partners in learning story structure while working through real decomposition problems.
The Pair Programming Approach
A working method I've found useful with GenAI on story writing mirrors
pair programming you act as the navigator while the AI drives. This means making small, iterative requests and approving each step before proceeding to the next level of detail.
Start with basic decomposition, get approval, then refine acceptance criteria. Ask for clarification on one aspect, review the response, then build on that foundation. This iterative approach serves two purposes: it keeps the GenAI focused on your specific context and requirements, and it builds your understanding of story structure progressively.
The key is maintaining control of the direction while letting the GenAI handle the vocabulary and formatting. You know what the system needs to do; the GenAI helps translate that knowledge into well-structured stories with appropriate acceptance criteria and clear boundaries.
Building Your Story Vocabulary with ChatGPT
ChatGPT works well for learning the fundamentals of story structure. Start with basic questions to establish the vocabulary: "What's the difference between an epic and a story?" or "What should acceptance criteria include for a user registration story?"
These foundational conversations build the terminology you'll use in more detailed prompts. Instead of asking for "user login stuff," you learn to request "user authentication stories with acceptance criteria separated by frontend and backend concerns."
The iterative approach helps here too. Ask ChatGPT to explain a concept, then work through it with your specific epic. "Now show me how to split my User Auth epic using those principles." This creates a feedback loop between learning the vocabulary and applying it to real work.
Once you've established the basic structure and terminology with ChatGPT, you can use that foundation as input for more sophisticated refinement with other tools.
A couple of practical considerations: GenAI tools don't retain memory between sessions, so they re-read your entire conversation history each time. Long conversations can lose focus as the context grows. When you feel you've covered enough ground on vocabulary and basic structure, ask the GenAI to prepare a summary prompt based on your current chat. This gives you a clean starting point for the next phase of work without carrying forward unnecessary context.
Claude as Your Story Review Partner
While ChatGPT helps establish the fundamentals, Claude works well for iterative refinement and review. Take the basic story structure you've developed and use Claude to examine specific aspects: rules clarification, interface separation, and quality review.
Claude handles the iterative refinement process effectively. You can ask it to review a story for splitting opportunities, suggest better acceptance criteria, or identify where frontend and backend concerns might be better separated. The pair programming approach continues here - you guide the direction while Claude provides detailed analysis and suggestions.
One useful pattern is using Claude as a second opinion on ChatGPT's output. Feed Claude a story that ChatGPT helped create and ask for review: "Does this story have clear boundaries? Are there splitting opportunities I'm missing?" This cross-validation helps catch issues that might not be obvious when working with a single tool.
Scope balance matters when asking for reviews. Instead of "Change my Frontend task," provide broader context: "Review this user registration story for splitting opportunities." GenAI works better with enough context to understand the full picture, but not so much that it loses focus. Finding this balance takes practice, but it's worth learning because it applies to all GenAI interactions.
The rules clarification aspect works particularly well with Claude. It can help identify acceptance criteria that are too vague, suggest more specific conditions, and point out edge cases that should be addressed before implementation.
The Breakthrough: Epic → Story → Task Structure
The most valuable outcome of working with GenAI on story writing isn't the individual stories themselves, but developing a clear mental model of how epics decompose into implementable work. This creates a 10,000-foot view - you start seeing the logical structure that connects high-level business goals to specific development tasks.
Working through the User Auth epic illustrates this progression. The original epic bundled everything together:
**Epic 1: User Auth & Profiles for marketplace**
*Implement user registration, login, authentication, session handling, and basic profile management.*
Through iterative work with GenAI, this decomposes into focused stories:
- User Registration & Login
- Seller Registration & Login
- Authentication (email/password)
- Session management
- Basic user profiles (name, email, preferences)
Each story then breaks down into specific tasks with clear acceptance criteria, separating front-end concerns from back-end implementation, and identifying the data boundaries and business rules that need to be addressed.
This structure becomes a reusable pattern. Once you've seen how authentication epics decompose, you recognize similar patterns in other domains and can apply the same decomposition techniques more confidently.
Learning While Doing
The iterative approach with GenAI builds story-writing skills alongside completing actual work. Each conversation teaches vocabulary, decomposition patterns, and quality criteria that apply to future stories.
The learning compounds over time. Early conversations focus on basic vocabulary and structure. As your prompting improves, you can tackle more complex decomposition challenges and start recognizing when AI suggestions don't align with your technical understanding. You learn to accept good suggestions while catching problematic ones.
The language barrier that initially makes prompting difficult can become an advantage through practice. Building precise vocabulary for story writing improves your ability to create effective prompts and work more efficiently with GenAI tools.
Rather than outsourcing story writing to GenAI, you're using GenAI as a companion to extend your knowledge and improve your work with stories.
Practical Next Steps
For the start, begin with a simple epic from your current work. Use ChatGPT to establish basic vocabulary - ask it to explain the difference between epics and stories, then try applying that understanding to your specific epic.
Work iteratively with small requests. Instead of asking for "complete stories with all details," start with "help me split this epic into 3-4 focused stories." Review the results, approve the direction, then ask for more detail on acceptance criteria or task breakdown.
When you feel comfortable with the basic structure, move to Claude for refinement. Use it to review your stories for splitting opportunities or to clarify acceptance criteria that seem too vague.
The free tiers of both ChatGPT and Claude provide enough capacity to work through creating one complete story, so you can experiment with this approach without any upfront cost. This gives you a chance to develop your working rhythm and see if the approach fits your needs before considering paid plans.
Remember that the goal isn't perfect stories on the first try, but building your understanding of story structure while completing real work. The stories improve as your prompting skills develop, and both contribute to better requirements management for your team.