In 30 seconds
- Keep the user and value; add precise context and checks.
- Acceptance criteria become executable tests.
- The conversation still happens, between humans, before the agent starts.
What to add
- Relevant files, modules or interfaces
- Coding conventions and patterns to follow
- Explicit acceptance criteria, ideally as tests
- Constraints and things not to change
- How to verify: commands to run, checks to pass
A simple template
As a [user], I want [goal], so that [value].
Context: where in the code. Done when: these tests pass. Don't: change the public API.
Keeping the 3 Cs
The card and confirmation get more precise; the conversation still matters. Humans should agree intent in refinement before an agent starts. Check stories with INVEST; Small and Testable matter even more for agents.
Myth vs reality
Myth · Agents need long, detailed stories.
Reality · They need precise ones; length isn't precision.
Myth · The template no longer matters.
Reality · The user and value still guide good decisions.
Myth · Agents can refine their own stories.
Reality · They can suggest; humans decide.
FAQ
Should every story go to an agent?
No; ambiguous or high-risk work suits humans first.
What if the agent misunderstands?
Improve the story, not just the output.
Do agents need the Definition of Done?
Yes, in executable form where possible.