structured output
technique judged 72% likely a named technique or architecture
See how it has spread across media tiers
Mentions over time
Table view
| Week | Articles | Sources |
|---|---|---|
| 2024-W38 | 1 | 1 |
| 2025-W21 | 1 | 1 |
| 2026-W01 | 1 | 1 |
| 2026-W02 | 1 | 1 |
| 2026-W09 | 1 | 1 |
| 2026-W11 | 13 | 1 |
| 2026-W24 | 1 | 1 |
| 2026-W25 | 2 | 2 |
| 2026-W26 | 1 | 1 |
| 2026-W29 | 1 | 1 |
| 2026-W30 | 1 | 1 |
| 2026-W34 | 2 | 2 |
| 2026-W35 | 3 | 3 |
| 2026-W36 | 1 | 1 |
| 2026-W39 | 1 | 1 |
| 2026-W40 | 18 | 3 |
| 2026-W41 | 6 | 2 |
In their words
Instead, maintain explicit state: class AgentState : user_id : str current_intent : SupportIntent conversation_id : str extracted_entities : dict # {"amount": 50000, "product": "laptop"} context_retrieved : dict # what we fetched and when decision_history : list # prior actions and outcomes timestamp : float def update_state ( state : AgentState , llm_response : str , user_input : str ): # Extract
ORDER BY version", (name,) ) return [self._row_to_prompt(r) for r in cursor.fetchall()] def _row_to_prompt(self, row) -> PromptVersion: return PromptVersion( id=row[0], name=row[1], version=row[2], prompt=row[3], metadata=json.loads(row[4]), created_at=row[5], parent_id=row[6] ) ### Prompt Deployment with Feature Flags python prompting/deployment.py from dataclasses import dataclass from typing: D
lower () for w in [ " why " , " how " , " compare " , " analyze " , " explain " , " pros " , " cons " ]) if word_count < 10 and not needs_reasoning : return " simple " elif needs_reasoning or has_multiple_questions or word_count > 30 : return " complex " return " medium " Complete Production Checklist Ingestion [ ] Semantic chunking with header preservation [ ] Metadata enrichment (source, date, d
with_structured_output ( EvaluationResult ) def evaluate_response ( task : str , input_text : str , response : str , criteria : list [ str ]) -> EvaluationResult : criteria_text = " \n " .
""" Model-Specific Tips GPT-4 / Claude Excellent at following complex instructions Handles long context well Good at structured output (JSON, YAML) Prefers explicit constraints Llama / Mistral (Open Weights) Benefits from more examples Needs clearer step-by-step Sometimes ignores negative constraints ("don't do X") Better with XML tags than markdown Small Models (7B and below) Few-shot critical Ke
Map each candidate to the kind of specialist model that fits it: a small generative model or a trained classifier for page types, a zero-shot entity model such as GLiNER2 or NuNER_Zero for entity tagging, a cross-encoder reranker in place of an LLM asked "is this chunk relevant?", and a document model plus a structured-output model for table fields.
The cost is more code and a more structured output from the model.
State invariants are still enforceable, only more expensive: validation steps, structured outputs, termination conditions that halt the team on violation.