GET /agents/runs/{id}, error.type = "AgentLoopStructuredOutputRetriesExhausted"
When this error occurs
When the model emits a final structured answer, the platform validates it against the run’soutput_format_schema. A validation failure no longer fails the run outright — instead the
workflow feeds the failure back to the model (the non-conforming output plus the validator’s
explanation) and asks it to try again. This error means every one of those retries also failed
validation, so the workflow gave up.
- The retry budget is a fixed 10 attempts — separate from
max_iterations, which bounds productive tool-use rounds rather than these corrective re-prompts. - Each retry is one inference job, so a run that hits this error has spent 11 inference jobs on the final answer alone (the first attempt plus 10 retries).
- Distinct from Output Schema Decode Failed: that one is a single, non-retried decode failure of the final answer; this one is the terminal state after the retry loop is exhausted.
How to fix
- Simplify the schema — Bedrock’s structured-output sampler enforces a subset of Draft
2020-12; deeply nested or exotic constructs (e.g. unconstrained
oneOf) are the most common reason the model can’t produce a conforming object even when re-prompted. - Reinforce the schema in the system prompt — tell the model explicitly what shape to emit and which fields are required. A schema the sampler struggles with is often one the model was never steered toward in the instructions.
- Inspect the last
failed_structured_outputon the final inference job’s result — it holds the raw JSON the model produced and the validator’s message, which pinpoints the field that keeps failing.

