What This Looks Like
The AI produces an answer, file, field, payload, or structured response that appears usable to the user, but the next step cannot consume it. A parser may reject it, a workflow may stop, a tool may fail, or a reviewer may be unable to continue because the output does not meet the next consumer’s requirements.
Why It Matters
AI output often sits in a chain. If the next step cannot use it, the answer is not operationally complete even if it looks reasonable in isolation. This can create hidden breakage where the user only discovers the issue after trying to pass the output into another system.
Structural Signal
The output crosses one boundary successfully but fails at the next interface. The issue is not only whether the output is understandable; it is whether the output fits the declared requirements of the downstream consumer.
Common Triggers
- The prompt optimizes for readable output instead of downstream compatibility
- The next step expects stricter structure than the user-facing answer provides
- Required fields, formats, labels, or delimiters are not preserved
- The answer mixes explanation with machine-consumable output
- The tool or workflow consumer has constraints not visible in the prompt
- A valid-looking answer violates a parser, importer, schema, or handoff contract
When to Use This Issue
Use this Issue when the AI output fails because the next workflow step, tool, parser, importer, or downstream consumer cannot use it as produced.
When Not to Use This Issue
Do not use this Issue when the output is simply factually wrong, incomplete for the user, or poorly written. Use it when the failure appears at the handoff into the next step.