Acceptance Criteria
Acceptance criteria are the explicit conditions a feature must meet before you consider it finished. They define exactly what "done" looks like before the work even starts.
Why acceptance criteria matter for AI coding agents
Ask an unconstrained agent to "implement user login," and it will likely hand you a working form with zero email validation, no error handling, and no secure redirect. Technically, it did what you asked. Functionally, it is useless.
Acceptance criteria close this gap. They force the AI to build to your standard, not its own. They dictate exactly what inputs are required, what the UI shows when a server fails, and what state the database must be in after the task is complete. More importantly, they give you a manual checklist. If the acceptance criteria are clear, you know exactly what to test before you clear the agent to start the next feature.
What acceptance criteria typically include
- The happy path: What the user should experience when everything works perfectly.
- The error states: What the UI shows when required inputs are missing or invalid.
- The fallback behavior: What happens when an external service or API goes down.
- The negative constraints: What the agent is explicitly forbidden from changing while building the feature.
For example, acceptance criteria for a new trip creation feature shouldn't just be "user can make a trip." It should specify: 1. A signed-in user can submit a destination, date, and budget. 2. Missing fields trigger a red inline error message. 3. The data saves cleanly to the trips table. 4. The user is instantly routed to the trip detail page upon success.
Zalcro generates dependency-aware tickets with built-in acceptance criteria. Your agent knows exactly what "done" looks like before it even starts the task.