Example (Zeeguu): Google Translate serves as the primary translation engine. When a user indicates the translation is inadequate, the system escalates to an LLM for a more nuanced, context-aware translation. This keeps costs low and speed high in the common case while providing higher, LLM-quality results when needed.
Forces: Specialized tools (translation APIs, NLP pipelines, classical classifiers) are faster, cheaper, and more deterministic than LLMs for well-defined tasks — but they sometimes fail or produce insufficiently satisfactory results.
Solution: Use the specialized tool as the primary path and escalate to the LLM only when the primary fails or the user signals dissatisfaction.
Applicability: This pattern applies broadly — topic classification, named entity recognition, or any NLP task where a cheaper tool handles the common case and the LLM handles the long tail.